Using cpu device Logging to runs/ppo_debug/ppo_1 ------------------------------ | time/ | | | fps | 5496 | | iterations | 1 | | time_elapsed | 2 | | total_timesteps | 16384 | ------------------------------ ------------------------------------------ | time/ | | | fps | 4317 | | iterations | 2 | | time_elapsed | 7 | | total_timesteps | 32768 | | train/ | | | approx_kl | 0.0036917897 | | clip_fraction | 0.0212 | | clip_range | 0.2 | | entropy_loss | -2.83 | | explained_variance | 0.352 | | learning_rate | 0.0003 | | loss | -0.0118 | | n_updates | 10 | | policy_gradient_loss | -0.000544 | | std | 0.999 | | value_loss | 0.0658 | ------------------------------------------ ------------------------------------------ | time/ | | | fps | 3946 | | iterations | 3 | | time_elapsed | 12 | | total_timesteps | 49152 | | train/ | | | approx_kl | 0.0033213054 | | clip_fraction | 0.0266 | | clip_range | 0.2 | | entropy_loss | -2.83 | | explained_variance | 0.502 | | learning_rate | 0.0003 | | loss | -0.0255 | | n_updates | 20 | | policy_gradient_loss | -0.00158 | | std | 0.997 | | value_loss | 0.08 | ------------------------------------------ /home/jalf/miniconda3/envs/tir/lib/python3.12/site-packages/stable_baselines3/common/evaluation.py:71: UserWarning: Evaluation environment is not wrapped with a ``Monitor`` wrapper. This may result in reporting modified episode lengths and rewards, if other wrappers happen to modify these. Consider wrapping environment first with ``Monitor`` wrapper. warnings.warn( Eval num_timesteps=50000, episode_reward=-32.92 +/- 15.12 Episode length: 2000.00 +/- 0.00 ----------------------------------------- | eval/ | | | mean_ep_length | 2e+03 | | mean_reward | -32.9 | | time/ | | | total_timesteps | 50000 | | train/ | | | approx_kl | 0.005147726 | | clip_fraction | 0.0478 | | clip_range | 0.2 | | entropy_loss | -2.84 | | explained_variance | 0.893 | | learning_rate | 0.0003 | | loss | -0.0145 | | n_updates | 30 | | policy_gradient_loss | -0.00318 | | std | 1 | | value_loss | 0.0194 | ----------------------------------------- New best mean reward! ------------------------------ | time/ | | | fps | 2231 | | iterations | 4 | | time_elapsed | 29 | | total_timesteps | 65536 | ------------------------------ ------------------------------------------ | time/ | | | fps | 2444 | | iterations | 5 | | time_elapsed | 33 | | total_timesteps | 81920 | | train/ | | | approx_kl | 0.0054671075 | | clip_fraction | 0.0529 | | clip_range | 0.2 | | entropy_loss | -2.84 | | explained_variance | 0.914 | | learning_rate | 0.0003 | | loss | -0.021 | | n_updates | 40 | | policy_gradient_loss | -0.00416 | | std | 1 | | value_loss | 0.0247 | ------------------------------------------ ----------------------------------------- | time/ | | | fps | 2616 | | iterations | 6 | | time_elapsed | 37 | | total_timesteps | 98304 | | train/ | | | approx_kl | 0.004603466 | | clip_fraction | 0.0379 | | clip_range | 0.2 | | entropy_loss | -2.83 | | explained_variance | 0.931 | | learning_rate | 0.0003 | | loss | -0.0193 | | n_updates | 50 | | policy_gradient_loss | -0.00284 | | std | 0.995 | | value_loss | 0.0171 | ----------------------------------------- /home/jalf/miniconda3/envs/tir/lib/python3.12/site-packages/stable_baselines3/common/evaluation.py:71: UserWarning: Evaluation environment is not wrapped with a ``Monitor`` wrapper. This may result in reporting modified episode lengths and rewards, if other wrappers happen to modify these. Consider wrapping environment first with ``Monitor`` wrapper. warnings.warn( Eval num_timesteps=100000, episode_reward=-27.45 +/- 49.10 Episode length: 1973.15 +/- 86.14 ------------------------------------------ | eval/ | | | mean_ep_length | 1.97e+03 | | mean_reward | -27.4 | | time/ | | | total_timesteps | 100000 | | train/ | | | approx_kl | 0.0053039393 | | clip_fraction | 0.0564 | | clip_range | 0.2 | | entropy_loss | -2.83 | | explained_variance | 0.878 | | learning_rate | 0.0003 | | loss | -0.0325 | | n_updates | 60 | | policy_gradient_loss | -0.00404 | | std | 0.998 | | value_loss | 0.0118 | ------------------------------------------ New best mean reward! ------------------------------- | time/ | | | fps | 2212 | | iterations | 7 | | time_elapsed | 51 | | total_timesteps | 114688 | ------------------------------- ------------------------------------------ | time/ | | | fps | 2332 | | iterations | 8 | | time_elapsed | 56 | | total_timesteps | 131072 | | train/ | | | approx_kl | 0.0048020086 | | clip_fraction | 0.0449 | | clip_range | 0.2 | | entropy_loss | -2.84 | | explained_variance | 0.839 | | learning_rate | 0.0003 | | loss | -0.0375 | | n_updates | 70 | | policy_gradient_loss | -0.00359 | | std | 1 | | value_loss | 0.0102 | ------------------------------------------ ----------------------------------------- | time/ | | | fps | 2429 | | iterations | 9 | | time_elapsed | 60 | | total_timesteps | 147456 | | train/ | | | approx_kl | 0.004460754 | | clip_fraction | 0.0349 | | clip_range | 0.2 | | entropy_loss | -2.85 | | explained_variance | 0.874 | | learning_rate | 0.0003 | | loss | -0.0293 | | n_updates | 80 | | policy_gradient_loss | -0.00294 | | std | 1.01 | | value_loss | 0.0132 | ----------------------------------------- Eval num_timesteps=150000, episode_reward=-33.46 +/- 39.53 Episode length: 1990.60 +/- 40.97 ----------------------------------------- | eval/ | | | mean_ep_length | 1.99e+03 | | mean_reward | -33.5 | | time/ | | | total_timesteps | 150000 | | train/ | | | approx_kl | 0.003831089 | | clip_fraction | 0.0196 | | clip_range | 0.2 | | entropy_loss | -2.82 | | explained_variance | 0.381 | | learning_rate | 0.0003 | | loss | -0.0191 | | n_updates | 90 | | policy_gradient_loss | -0.00202 | | std | 0.984 | | value_loss | 0.104 | ----------------------------------------- ------------------------------- | time/ | | | fps | 2147 | | iterations | 10 | | time_elapsed | 76 | | total_timesteps | 163840 | ------------------------------- ------------------------------------------ | time/ | | | fps | 2218 | | iterations | 11 | | time_elapsed | 81 | | total_timesteps | 180224 | | train/ | | | approx_kl | 0.0032510734 | | clip_fraction | 0.0246 | | clip_range | 0.2 | | entropy_loss | -2.82 | | explained_variance | 0.887 | | learning_rate | 0.0003 | | loss | -0.0279 | | n_updates | 100 | | policy_gradient_loss | -0.00207 | | std | 0.993 | | value_loss | 0.045 | ------------------------------------------ ------------------------------------------ | time/ | | | fps | 2289 | | iterations | 12 | | time_elapsed | 85 | | total_timesteps | 196608 | | train/ | | | approx_kl | 0.0047060847 | | clip_fraction | 0.0387 | | clip_range | 0.2 | | entropy_loss | -2.83 | | explained_variance | 0.896 | | learning_rate | 0.0003 | | loss | 0.00931 | | n_updates | 110 | | policy_gradient_loss | -0.00305 | | std | 0.994 | | value_loss | 0.0489 | ------------------------------------------ Eval num_timesteps=200000, episode_reward=-18.47 +/- 55.53 Episode length: 1938.95 +/- 147.97 ------------------------------------------ | eval/ | | | mean_ep_length | 1.94e+03 | | mean_reward | -18.5 | | time/ | | | total_timesteps | 200000 | | train/ | | | approx_kl | 0.0047602034 | | clip_fraction | 0.0421 | | clip_range | 0.2 | | entropy_loss | -2.86 | | explained_variance | 0.968 | | learning_rate | 0.0003 | | loss | -0.0301 | | n_updates | 120 | | policy_gradient_loss | -0.00281 | | std | 1.01 | | value_loss | 0.0094 | ------------------------------------------ New best mean reward! [Diag @ 200,000 | n_sheep=1 | success=5%] COMPACT_CANT_DRIVE 18/20 DROVE_NO_SHEEP 1/20 SUCCESS 1/20 action_mag mean=0.269 p10=0.129 p90=0.447 (0=stopped, 1=full speed) min_flock_radius mean=0.00m best=0.00m (target <5m to compact) min_dog_to_com mean=3.86m best=1.91m (FLEE_DIST=7m) min_com_to_pen mean=11.22m best=2.44m reward/step (mean): progress=-0.0022 alignment=+0.0006 pen_bonus=+0.0003 step_cost=-0.0200 complete=+0.0026 ------------------------------- | time/ | | | fps | 1964 | | iterations | 13 | | time_elapsed | 108 | | total_timesteps | 212992 | ------------------------------- ------------------------------------------ | time/ | | | fps | 2034 | | iterations | 14 | | time_elapsed | 112 | | total_timesteps | 229376 | | train/ | | | approx_kl | 0.0041663316 | | clip_fraction | 0.0373 | | clip_range | 0.2 | | entropy_loss | -2.88 | | explained_variance | 0.901 | | learning_rate | 0.0003 | | loss | -0.0251 | | n_updates | 130 | | policy_gradient_loss | -0.00223 | | std | 1.03 | | value_loss | 0.00752 | ------------------------------------------ ------------------------------------------ | time/ | | | fps | 2102 | | iterations | 15 | | time_elapsed | 116 | | total_timesteps | 245760 | | train/ | | | approx_kl | 0.0042076977 | | clip_fraction | 0.032 | | clip_range | 0.2 | | entropy_loss | -2.91 | | explained_variance | 0.939 | | learning_rate | 0.0003 | | loss | -0.0333 | | n_updates | 140 | | policy_gradient_loss | -0.00281 | | std | 1.04 | | value_loss | 0.00934 | ------------------------------------------ Eval num_timesteps=250000, episode_reward=-37.07 +/- 35.02 Episode length: 1938.20 +/- 269.38 ------------------------------------------ | eval/ | | | mean_ep_length | 1.94e+03 | | mean_reward | -37.1 | | time/ | | | total_timesteps | 250000 | | train/ | | | approx_kl | 0.0028561926 | | clip_fraction | 0.0171 | | clip_range | 0.2 | | entropy_loss | -2.92 | | explained_variance | 0.822 | | learning_rate | 0.0003 | | loss | -0.0292 | | n_updates | 150 | | policy_gradient_loss | -0.00113 | | std | 1.04 | | value_loss | 0.0473 | ------------------------------------------ ------------------------------- | time/ | | | fps | 1990 | | iterations | 16 | | time_elapsed | 131 | | total_timesteps | 262144 | ------------------------------- ------------------------------------------ | time/ | | | fps | 2042 | | iterations | 17 | | time_elapsed | 136 | | total_timesteps | 278528 | | train/ | | | approx_kl | 0.0054259067 | | clip_fraction | 0.0468 | | clip_range | 0.2 | | entropy_loss | -2.91 | | explained_variance | 0.891 | | learning_rate | 0.0003 | | loss | -0.032 | | n_updates | 160 | | policy_gradient_loss | -0.00597 | | std | 1.03 | | value_loss | 0.0128 | ------------------------------------------ ----------------------------------------- | time/ | | | fps | 2085 | | iterations | 18 | | time_elapsed | 141 | | total_timesteps | 294912 | | train/ | | | approx_kl | 0.004205579 | | clip_fraction | 0.0291 | | clip_range | 0.2 | | entropy_loss | -2.91 | | explained_variance | 0.834 | | learning_rate | 0.0003 | | loss | -0.0364 | | n_updates | 170 | | policy_gradient_loss | -0.00307 | | std | 1.03 | | value_loss | 0.0107 | ----------------------------------------- Eval num_timesteps=300000, episode_reward=-25.41 +/- 48.70 Episode length: 1886.45 +/- 435.99 ------------------------------------------ | eval/ | | | mean_ep_length | 1.89e+03 | | mean_reward | -25.4 | | time/ | | | total_timesteps | 300000 | | train/ | | | approx_kl | 0.0045948992 | | clip_fraction | 0.0354 | | clip_range | 0.2 | | entropy_loss | -2.9 | | explained_variance | 0.806 | | learning_rate | 0.0003 | | loss | -0.0242 | | n_updates | 180 | | policy_gradient_loss | -0.00236 | | std | 1.03 | | value_loss | 0.0371 | ------------------------------------------ ------------------------------- | time/ | | | fps | 1981 | | iterations | 19 | | time_elapsed | 157 | | total_timesteps | 311296 | ------------------------------- ----------------------------------------- | time/ | | | fps | 2024 | | iterations | 20 | | time_elapsed | 161 | | total_timesteps | 327680 | | train/ | | | approx_kl | 0.005344864 | | clip_fraction | 0.0442 | | clip_range | 0.2 | | entropy_loss | -2.91 | | explained_variance | 0.877 | | learning_rate | 0.0003 | | loss | -0.0369 | | n_updates | 190 | | policy_gradient_loss | -0.00344 | | std | 1.04 | | value_loss | 0.0104 | ----------------------------------------- ----------------------------------------- | time/ | | | fps | 2066 | | iterations | 21 | | time_elapsed | 166 | | total_timesteps | 344064 | | train/ | | | approx_kl | 0.007574372 | | clip_fraction | 0.0753 | | clip_range | 0.2 | | entropy_loss | -2.92 | | explained_variance | 0.903 | | learning_rate | 0.0003 | | loss | -0.0272 | | n_updates | 200 | | policy_gradient_loss | -0.00726 | | std | 1.04 | | value_loss | 0.0113 | ----------------------------------------- Eval num_timesteps=350000, episode_reward=-21.14 +/- 37.01 Episode length: 1959.80 +/- 175.23 ------------------------------------------ | eval/ | | | mean_ep_length | 1.96e+03 | | mean_reward | -21.1 | | time/ | | | total_timesteps | 350000 | | train/ | | | approx_kl | 0.0061714016 | | clip_fraction | 0.0569 | | clip_range | 0.2 | | entropy_loss | -2.91 | | explained_variance | 0.917 | | learning_rate | 0.0003 | | loss | -0.022 | | n_updates | 210 | | policy_gradient_loss | -0.00598 | | std | 1.04 | | value_loss | 0.0231 | ------------------------------------------ ------------------------------- | time/ | | | fps | 1984 | | iterations | 22 | | time_elapsed | 181 | | total_timesteps | 360448 | ------------------------------- ---------------------------------------- | time/ | | | fps | 2026 | | iterations | 23 | | time_elapsed | 185 | | total_timesteps | 376832 | | train/ | | | approx_kl | 0.00587913 | | clip_fraction | 0.0501 | | clip_range | 0.2 | | entropy_loss | -2.92 | | explained_variance | 0.932 | | learning_rate | 0.0003 | | loss | -0.0415 | | n_updates | 220 | | policy_gradient_loss | -0.00484 | | std | 1.04 | | value_loss | 0.0242 | ---------------------------------------- ----------------------------------------- | time/ | | | fps | 2064 | | iterations | 24 | | time_elapsed | 190 | | total_timesteps | 393216 | | train/ | | | approx_kl | 0.006933649 | | clip_fraction | 0.081 | | clip_range | 0.2 | | entropy_loss | -2.91 | | explained_variance | 0.918 | | learning_rate | 0.0003 | | loss | -0.032 | | n_updates | 230 | | policy_gradient_loss | -0.00773 | | std | 1.03 | | value_loss | 0.0233 | ----------------------------------------- Eval num_timesteps=400000, episode_reward=-2.75 +/- 37.08 Episode length: 1998.55 +/- 6.32 ------------------------------------------ | eval/ | | | mean_ep_length | 2e+03 | | mean_reward | -2.75 | | time/ | | | total_timesteps | 400000 | | train/ | | | approx_kl | 0.0064436095 | | clip_fraction | 0.0647 | | clip_range | 0.2 | | entropy_loss | -2.9 | | explained_variance | 0.853 | | learning_rate | 0.0003 | | loss | 0.0633 | | n_updates | 240 | | policy_gradient_loss | -0.00551 | | std | 1.03 | | value_loss | 0.128 | ------------------------------------------ New best mean reward! [Diag @ 400,000 | n_sheep=1 | success=0%] DROVE_NO_SHEEP 13/20 COMPACT_CANT_DRIVE 7/20 action_mag mean=0.316 p10=0.057 p90=0.512 (0=stopped, 1=full speed) min_flock_radius mean=0.00m best=0.00m (target <5m to compact) min_dog_to_com mean=1.86m best=0.95m (FLEE_DIST=7m) min_com_to_pen mean=3.19m best=1.50m reward/step (mean): progress=+0.0093 alignment=+0.0040 pen_bonus=+0.0000 step_cost=-0.0200 complete=+0.0000 ------------------------------- | time/ | | | fps | 1925 | | iterations | 25 | | time_elapsed | 212 | | total_timesteps | 409600 | ------------------------------- ---------------------------------------- | time/ | | | fps | 1961 | | iterations | 26 | | time_elapsed | 217 | | total_timesteps | 425984 | | train/ | | | approx_kl | 0.00806847 | | clip_fraction | 0.1 | | clip_range | 0.2 | | entropy_loss | -2.88 | | explained_variance | 0.933 | | learning_rate | 0.0003 | | loss | -0.0254 | | n_updates | 250 | | policy_gradient_loss | -0.00871 | | std | 1.02 | | value_loss | 0.0264 | ---------------------------------------- ----------------------------------------- | time/ | | | fps | 1997 | | iterations | 27 | | time_elapsed | 221 | | total_timesteps | 442368 | | train/ | | | approx_kl | 0.005784355 | | clip_fraction | 0.0531 | | clip_range | 0.2 | | entropy_loss | -2.86 | | explained_variance | 0.878 | | learning_rate | 0.0003 | | loss | 0.00996 | | n_updates | 260 | | policy_gradient_loss | -0.00485 | | std | 1 | | value_loss | 0.0868 | ----------------------------------------- Eval num_timesteps=450000, episode_reward=51.79 +/- 20.61 Episode length: 1912.30 +/- 382.28 ----------------------------------------- | eval/ | | | mean_ep_length | 1.91e+03 | | mean_reward | 51.8 | | time/ | | | total_timesteps | 450000 | | train/ | | | approx_kl | 0.005881632 | | clip_fraction | 0.0639 | | clip_range | 0.2 | | entropy_loss | -2.83 | | explained_variance | 0.952 | | learning_rate | 0.0003 | | loss | -0.0187 | | n_updates | 270 | | policy_gradient_loss | -0.00655 | | std | 0.991 | | value_loss | 0.0226 | ----------------------------------------- New best mean reward! ------------------------------- | time/ | | | fps | 1936 | | iterations | 28 | | time_elapsed | 236 | | total_timesteps | 458752 | ------------------------------- ----------------------------------------- | time/ | | | fps | 1965 | | iterations | 29 | | time_elapsed | 241 | | total_timesteps | 475136 | | train/ | | | approx_kl | 0.009020726 | | clip_fraction | 0.0982 | | clip_range | 0.2 | | entropy_loss | -2.81 | | explained_variance | 0.87 | | learning_rate | 0.0003 | | loss | 0.0218 | | n_updates | 280 | | policy_gradient_loss | -0.0061 | | std | 0.984 | | value_loss | 0.209 | ----------------------------------------- ----------------------------------------- | time/ | | | fps | 1999 | | iterations | 30 | | time_elapsed | 245 | | total_timesteps | 491520 | | train/ | | | approx_kl | 0.011525536 | | clip_fraction | 0.136 | | clip_range | 0.2 | | entropy_loss | -2.79 | | explained_variance | 0.92 | | learning_rate | 0.0003 | | loss | 0.0306 | | n_updates | 290 | | policy_gradient_loss | -0.00896 | | std | 0.97 | | value_loss | 0.0903 | ----------------------------------------- Eval num_timesteps=500000, episode_reward=87.01 +/- 42.12 Episode length: 1359.85 +/- 815.95 ----------------------------------------- | eval/ | | | mean_ep_length | 1.36e+03 | | mean_reward | 87 | | time/ | | | total_timesteps | 500000 | | train/ | | | approx_kl | 0.012545023 | | clip_fraction | 0.171 | | clip_range | 0.2 | | entropy_loss | -2.78 | | explained_variance | 0.956 | | learning_rate | 0.0003 | | loss | -0.0369 | | n_updates | 300 | | policy_gradient_loss | -0.0069 | | std | 0.972 | | value_loss | 0.034 | ----------------------------------------- New best mean reward! ------------------------------- | time/ | | | fps | 1968 | | iterations | 31 | | time_elapsed | 258 | | total_timesteps | 507904 | ------------------------------- ----------------------------------------- | time/ | | | fps | 1996 | | iterations | 32 | | time_elapsed | 262 | | total_timesteps | 524288 | | train/ | | | approx_kl | 0.008305798 | | clip_fraction | 0.102 | | clip_range | 0.2 | | entropy_loss | -2.78 | | explained_variance | 0.975 | | learning_rate | 0.0003 | | loss | -0.0285 | | n_updates | 310 | | policy_gradient_loss | -0.00343 | | std | 0.972 | | value_loss | 0.0162 | ----------------------------------------- ------------------------------------------ | time/ | | | fps | 2021 | | iterations | 33 | | time_elapsed | 267 | | total_timesteps | 540672 | | train/ | | | approx_kl | 0.0074599315 | | clip_fraction | 0.0925 | | clip_range | 0.2 | | entropy_loss | -2.81 | | explained_variance | 0.976 | | learning_rate | 0.0003 | | loss | -0.0282 | | n_updates | 320 | | policy_gradient_loss | -0.0028 | | std | 0.989 | | value_loss | 0.0136 | ------------------------------------------ Eval num_timesteps=550000, episode_reward=113.42 +/- 48.33 Episode length: 926.05 +/- 792.99 ----------------------------------------- | eval/ | | | mean_ep_length | 926 | | mean_reward | 113 | | time/ | | | total_timesteps | 550000 | | train/ | | | approx_kl | 0.010888291 | | clip_fraction | 0.136 | | clip_range | 0.2 | | entropy_loss | -2.83 | | explained_variance | 0.981 | | learning_rate | 0.0003 | | loss | -0.0226 | | n_updates | 330 | | policy_gradient_loss | -0.00266 | | std | 1 | | value_loss | 0.00643 | ----------------------------------------- New best mean reward! ------------------------------- | time/ | | | fps | 2005 | | iterations | 34 | | time_elapsed | 277 | | total_timesteps | 557056 | ------------------------------- ----------------------------------------- | time/ | | | fps | 2030 | | iterations | 35 | | time_elapsed | 282 | | total_timesteps | 573440 | | train/ | | | approx_kl | 0.009418717 | | clip_fraction | 0.121 | | clip_range | 0.2 | | entropy_loss | -2.84 | | explained_variance | 0.975 | | learning_rate | 0.0003 | | loss | -0.0234 | | n_updates | 340 | | policy_gradient_loss | -0.00417 | | std | 1 | | value_loss | 0.0219 | ----------------------------------------- ----------------------------------------- | time/ | | | fps | 2054 | | iterations | 36 | | time_elapsed | 287 | | total_timesteps | 589824 | | train/ | | | approx_kl | 0.009153167 | | clip_fraction | 0.132 | | clip_range | 0.2 | | entropy_loss | -2.86 | | explained_variance | 0.972 | | learning_rate | 0.0003 | | loss | 0.00458 | | n_updates | 350 | | policy_gradient_loss | -0.00925 | | std | 1.01 | | value_loss | 0.0644 | ----------------------------------------- Eval num_timesteps=600000, episode_reward=142.43 +/- 15.10 Episode length: 292.00 +/- 114.85 ------------------------------------------ | eval/ | | | mean_ep_length | 292 | | mean_reward | 142 | | time/ | | | total_timesteps | 600000 | | train/ | | | approx_kl | 0.0073751104 | | clip_fraction | 0.0817 | | clip_range | 0.2 | | entropy_loss | -2.85 | | explained_variance | 0.967 | | learning_rate | 0.0003 | | loss | 0.0205 | | n_updates | 360 | | policy_gradient_loss | -0.0078 | | std | 1.01 | | value_loss | 0.0854 | ------------------------------------------ New best mean reward! [Diag @ 600,000 | n_sheep=1 | success=100%] SUCCESS 20/20 action_mag mean=0.339 p10=0.246 p90=0.609 (0=stopped, 1=full speed) min_flock_radius mean=0.00m best=0.00m (target <5m to compact) min_dog_to_com mean=1.68m best=0.23m (FLEE_DIST=7m) min_com_to_pen mean=3.54m best=2.70m reward/step (mean): progress=+0.0996 alignment=+0.0271 pen_bonus=+0.0302 step_cost=-0.0200 complete=+0.3022 ------------------------------- | time/ | | | fps | 2059 | | iterations | 37 | | time_elapsed | 294 | | total_timesteps | 606208 | ------------------------------- ----------------------------------------- | time/ | | | fps | 2069 | | iterations | 38 | | time_elapsed | 300 | | total_timesteps | 622592 | | train/ | | | approx_kl | 0.006348365 | | clip_fraction | 0.0685 | | clip_range | 0.2 | | entropy_loss | -2.85 | | explained_variance | 0.954 | | learning_rate | 0.0003 | | loss | -0.0107 | | n_updates | 370 | | policy_gradient_loss | -0.00403 | | std | 1 | | value_loss | 0.0629 | ----------------------------------------- ------------------------------------------ | time/ | | | fps | 2085 | | iterations | 39 | | time_elapsed | 306 | | total_timesteps | 638976 | | train/ | | | approx_kl | 0.0073653567 | | clip_fraction | 0.089 | | clip_range | 0.2 | | entropy_loss | -2.83 | | explained_variance | 0.976 | | learning_rate | 0.0003 | | loss | -0.0379 | | n_updates | 380 | | policy_gradient_loss | -0.00635 | | std | 0.993 | | value_loss | 0.0213 | ------------------------------------------ Eval num_timesteps=650000, episode_reward=148.63 +/- 11.08 Episode length: 312.15 +/- 83.52 ------------------------------------------ | eval/ | | | mean_ep_length | 312 | | mean_reward | 149 | | time/ | | | total_timesteps | 650000 | | train/ | | | approx_kl | 0.0064217458 | | clip_fraction | 0.0662 | | clip_range | 0.2 | | entropy_loss | -2.81 | | explained_variance | 0.977 | | learning_rate | 0.0003 | | loss | -0.0177 | | n_updates | 390 | | policy_gradient_loss | -0.00451 | | std | 0.983 | | value_loss | 0.0325 | ------------------------------------------ New best mean reward! ------------------------------- | time/ | | | fps | 2092 | | iterations | 40 | | time_elapsed | 313 | | total_timesteps | 655360 | ------------------------------- ----------------------------------------- | time/ | | | fps | 2107 | | iterations | 41 | | time_elapsed | 318 | | total_timesteps | 671744 | | train/ | | | approx_kl | 0.007330196 | | clip_fraction | 0.0823 | | clip_range | 0.2 | | entropy_loss | -2.79 | | explained_variance | 0.985 | | learning_rate | 0.0003 | | loss | -0.0257 | | n_updates | 400 | | policy_gradient_loss | -0.00559 | | std | 0.971 | | value_loss | 0.0108 | ----------------------------------------- ------------------------------------------ | time/ | | | fps | 2123 | | iterations | 42 | | time_elapsed | 323 | | total_timesteps | 688128 | | train/ | | | approx_kl | 0.0076610697 | | clip_fraction | 0.0876 | | clip_range | 0.2 | | entropy_loss | -2.77 | | explained_variance | 0.99 | | learning_rate | 0.0003 | | loss | -0.037 | | n_updates | 410 | | policy_gradient_loss | -0.00581 | | std | 0.966 | | value_loss | 0.00623 | ------------------------------------------ Eval num_timesteps=700000, episode_reward=137.38 +/- 18.54 Episode length: 255.10 +/- 119.47 ------------------------------------------ | eval/ | | | mean_ep_length | 255 | | mean_reward | 137 | | time/ | | | total_timesteps | 700000 | | train/ | | | approx_kl | 0.0072219693 | | clip_fraction | 0.0734 | | clip_range | 0.2 | | entropy_loss | -2.76 | | explained_variance | 0.989 | | learning_rate | 0.0003 | | loss | -0.0383 | | n_updates | 420 | | policy_gradient_loss | -0.00416 | | std | 0.961 | | value_loss | 0.00951 | ------------------------------------------ ------------------------------- | time/ | | | fps | 2128 | | iterations | 43 | | time_elapsed | 331 | | total_timesteps | 704512 | ------------------------------- ------------------------------------------ | time/ | | | fps | 2144 | | iterations | 44 | | time_elapsed | 336 | | total_timesteps | 720896 | | train/ | | | approx_kl | 0.0075956425 | | clip_fraction | 0.0895 | | clip_range | 0.2 | | entropy_loss | -2.75 | | explained_variance | 0.993 | | learning_rate | 0.0003 | | loss | -0.0433 | | n_updates | 430 | | policy_gradient_loss | -0.00475 | | std | 0.953 | | value_loss | 0.00343 | ------------------------------------------ ------------------------------------------ | time/ | | | fps | 2160 | | iterations | 45 | | time_elapsed | 341 | | total_timesteps | 737280 | | train/ | | | approx_kl | 0.0062526334 | | clip_fraction | 0.0699 | | clip_range | 0.2 | | entropy_loss | -2.72 | | explained_variance | 0.99 | | learning_rate | 0.0003 | | loss | -0.0329 | | n_updates | 440 | | policy_gradient_loss | -0.00355 | | std | 0.942 | | value_loss | 0.0113 | ------------------------------------------ Eval num_timesteps=750000, episode_reward=145.04 +/- 16.56 Episode length: 291.10 +/- 132.25 ------------------------------------------ | eval/ | | | mean_ep_length | 291 | | mean_reward | 145 | | time/ | | | total_timesteps | 750000 | | train/ | | | approx_kl | 0.0058749127 | | clip_fraction | 0.0607 | | clip_range | 0.2 | | entropy_loss | -2.71 | | explained_variance | 0.993 | | learning_rate | 0.0003 | | loss | -0.0281 | | n_updates | 450 | | policy_gradient_loss | -0.00324 | | std | 0.934 | | value_loss | 0.00811 | ------------------------------------------ ------------------------------- | time/ | | | fps | 2161 | | iterations | 46 | | time_elapsed | 348 | | total_timesteps | 753664 | ------------------------------- ------------------------------------------ | time/ | | | fps | 2176 | | iterations | 47 | | time_elapsed | 353 | | total_timesteps | 770048 | | train/ | | | approx_kl | 0.0070656985 | | clip_fraction | 0.0763 | | clip_range | 0.2 | | entropy_loss | -2.68 | | explained_variance | 0.996 | | learning_rate | 0.0003 | | loss | -0.0322 | | n_updates | 460 | | policy_gradient_loss | -0.00485 | | std | 0.92 | | value_loss | 0.00234 | ------------------------------------------ ----------------------------------------- | time/ | | | fps | 2193 | | iterations | 48 | | time_elapsed | 358 | | total_timesteps | 786432 | | train/ | | | approx_kl | 0.008987564 | | clip_fraction | 0.112 | | clip_range | 0.2 | | entropy_loss | -2.66 | | explained_variance | 0.997 | | learning_rate | 0.0003 | | loss | -0.0471 | | n_updates | 470 | | policy_gradient_loss | -0.00864 | | std | 0.909 | | value_loss | 0.00178 | ----------------------------------------- Eval num_timesteps=800000, episode_reward=141.03 +/- 13.75 Episode length: 256.90 +/- 100.39 ----------------------------------------- | eval/ | | | mean_ep_length | 257 | | mean_reward | 141 | | time/ | | | total_timesteps | 800000 | | train/ | | | approx_kl | 0.008297143 | | clip_fraction | 0.0945 | | clip_range | 0.2 | | entropy_loss | -2.67 | | explained_variance | 0.989 | | learning_rate | 0.0003 | | loss | -0.0173 | | n_updates | 480 | | policy_gradient_loss | -0.00352 | | std | 0.921 | | value_loss | 0.00934 | ----------------------------------------- [Diag @ 800,000 | n_sheep=1 | success=100%] SUCCESS 20/20 action_mag mean=0.333 p10=0.244 p90=0.332 (0=stopped, 1=full speed) min_flock_radius mean=0.00m best=0.00m (target <5m to compact) min_dog_to_com mean=1.40m best=0.75m (FLEE_DIST=7m) min_com_to_pen mean=3.47m best=1.58m reward/step (mean): progress=+0.1108 alignment=+0.0328 pen_bonus=+0.0366 step_cost=-0.0200 complete=+0.3664 [Curriculum] leaving stage n_sheep=1 after 800,000 steps | training success rate (last 100 eps) = 100% [Curriculum] → 2 sheep at step 800,000 ------------------------------- | time/ | | | fps | 2187 | | iterations | 49 | | time_elapsed | 367 | | total_timesteps | 802816 | ------------------------------- ----------------------------------------- | time/ | | | fps | 2201 | | iterations | 50 | | time_elapsed | 372 | | total_timesteps | 819200 | | train/ | | | approx_kl | 0.006534174 | | clip_fraction | 0.0754 | | clip_range | 0.2 | | entropy_loss | -2.7 | | explained_variance | 0.968 | | learning_rate | 0.0003 | | loss | -0.0252 | | n_updates | 490 | | policy_gradient_loss | 0.00248 | | std | 0.942 | | value_loss | 0.021 | ----------------------------------------- ----------------------------------------- | time/ | | | fps | 2213 | | iterations | 51 | | time_elapsed | 377 | | total_timesteps | 835584 | | train/ | | | approx_kl | 0.012509884 | | clip_fraction | 0.182 | | clip_range | 0.2 | | entropy_loss | -2.73 | | explained_variance | 0.51 | | learning_rate | 0.0003 | | loss | -0.0127 | | n_updates | 500 | | policy_gradient_loss | 0.00321 | | std | 0.953 | | value_loss | 0.0093 | ----------------------------------------- Eval num_timesteps=850000, episode_reward=-30.43 +/- 29.94 Episode length: 2000.00 +/- 0.00 ----------------------------------------- | eval/ | | | mean_ep_length | 2e+03 | | mean_reward | -30.4 | | time/ | | | total_timesteps | 850000 | | train/ | | | approx_kl | 0.009752454 | | clip_fraction | 0.146 | | clip_range | 0.2 | | entropy_loss | -2.74 | | explained_variance | 0.865 | | learning_rate | 0.0003 | | loss | -0.0289 | | n_updates | 510 | | policy_gradient_loss | 0.00274 | | std | 0.95 | | value_loss | 0.0117 | ----------------------------------------- ------------------------------- | time/ | | | fps | 2153 | | iterations | 52 | | time_elapsed | 395 | | total_timesteps | 851968 | ------------------------------- ----------------------------------------- | time/ | | | fps | 2166 | | iterations | 53 | | time_elapsed | 400 | | total_timesteps | 868352 | | train/ | | | approx_kl | 0.011746319 | | clip_fraction | 0.133 | | clip_range | 0.2 | | entropy_loss | -2.75 | | explained_variance | 0.953 | | learning_rate | 0.0003 | | loss | -0.0316 | | n_updates | 520 | | policy_gradient_loss | 0.00116 | | std | 0.958 | | value_loss | 0.00603 | ----------------------------------------- ----------------------------------------- | time/ | | | fps | 2179 | | iterations | 54 | | time_elapsed | 405 | | total_timesteps | 884736 | | train/ | | | approx_kl | 0.008340008 | | clip_fraction | 0.111 | | clip_range | 0.2 | | entropy_loss | -2.75 | | explained_variance | 0.959 | | learning_rate | 0.0003 | | loss | -0.0317 | | n_updates | 530 | | policy_gradient_loss | 0.000628 | | std | 0.955 | | value_loss | 0.00663 | ----------------------------------------- Eval num_timesteps=900000, episode_reward=-21.80 +/- 34.98 Episode length: 2000.00 +/- 0.00 ----------------------------------------- | eval/ | | | mean_ep_length | 2e+03 | | mean_reward | -21.8 | | time/ | | | total_timesteps | 900000 | | train/ | | | approx_kl | 0.010461532 | | clip_fraction | 0.13 | | clip_range | 0.2 | | entropy_loss | -2.74 | | explained_variance | 0.88 | | learning_rate | 0.0003 | | loss | -0.00905 | | n_updates | 540 | | policy_gradient_loss | -0.000256 | | std | 0.951 | | value_loss | 0.00567 | ----------------------------------------- ------------------------------- | time/ | | | fps | 2128 | | iterations | 55 | | time_elapsed | 423 | | total_timesteps | 901120 | ------------------------------- ------------------------------------------ | time/ | | | fps | 2139 | | iterations | 56 | | time_elapsed | 428 | | total_timesteps | 917504 | | train/ | | | approx_kl | 0.0071650296 | | clip_fraction | 0.0988 | | clip_range | 0.2 | | entropy_loss | -2.75 | | explained_variance | 0.931 | | learning_rate | 0.0003 | | loss | -0.0294 | | n_updates | 550 | | policy_gradient_loss | -0.000672 | | std | 0.957 | | value_loss | 0.00545 | ------------------------------------------ ----------------------------------------- | time/ | | | fps | 2152 | | iterations | 57 | | time_elapsed | 433 | | total_timesteps | 933888 | | train/ | | | approx_kl | 0.009678386 | | clip_fraction | 0.112 | | clip_range | 0.2 | | entropy_loss | -2.74 | | explained_variance | 0.927 | | learning_rate | 0.0003 | | loss | -0.0308 | | n_updates | 560 | | policy_gradient_loss | -0.000959 | | std | 0.953 | | value_loss | 0.00409 | ----------------------------------------- Eval num_timesteps=950000, episode_reward=-34.37 +/- 35.50 Episode length: 2000.00 +/- 0.00 ----------------------------------------- | eval/ | | | mean_ep_length | 2e+03 | | mean_reward | -34.4 | | time/ | | | total_timesteps | 950000 | | train/ | | | approx_kl | 0.008903094 | | clip_fraction | 0.111 | | clip_range | 0.2 | | entropy_loss | -2.74 | | explained_variance | 0.939 | | learning_rate | 0.0003 | | loss | -0.0259 | | n_updates | 570 | | policy_gradient_loss | -0.000299 | | std | 0.955 | | value_loss | 0.00432 | ----------------------------------------- ------------------------------- | time/ | | | fps | 2108 | | iterations | 58 | | time_elapsed | 450 | | total_timesteps | 950272 | ------------------------------- ----------------------------------------- | time/ | | | fps | 2117 | | iterations | 59 | | time_elapsed | 456 | | total_timesteps | 966656 | | train/ | | | approx_kl | 0.008592881 | | clip_fraction | 0.0954 | | clip_range | 0.2 | | entropy_loss | -2.74 | | explained_variance | 0.929 | | learning_rate | 0.0003 | | loss | -0.0173 | | n_updates | 580 | | policy_gradient_loss | 0.00103 | | std | 0.95 | | value_loss | 0.00265 | ----------------------------------------- ----------------------------------------- | time/ | | | fps | 2129 | | iterations | 60 | | time_elapsed | 461 | | total_timesteps | 983040 | | train/ | | | approx_kl | 0.010225108 | | clip_fraction | 0.108 | | clip_range | 0.2 | | entropy_loss | -2.74 | | explained_variance | 0.972 | | learning_rate | 0.0003 | | loss | -0.0135 | | n_updates | 590 | | policy_gradient_loss | -0.000738 | | std | 0.954 | | value_loss | 0.0029 | ----------------------------------------- ----------------------------------------- | time/ | | | fps | 2137 | | iterations | 61 | | time_elapsed | 467 | | total_timesteps | 999424 | | train/ | | | approx_kl | 0.008312117 | | clip_fraction | 0.0887 | | clip_range | 0.2 | | entropy_loss | -2.75 | | explained_variance | 0.898 | | learning_rate | 0.0003 | | loss | -0.0262 | | n_updates | 600 | | policy_gradient_loss | -0.000497 | | std | 0.958 | | value_loss | 0.00511 | ----------------------------------------- Eval num_timesteps=1000000, episode_reward=-32.64 +/- 38.38 Episode length: 2000.00 +/- 0.00 ---------------------------------------- | eval/ | | | mean_ep_length | 2e+03 | | mean_reward | -32.6 | | time/ | | | total_timesteps | 1000000 | | train/ | | | approx_kl | 0.00942917 | | clip_fraction | 0.105 | | clip_range | 0.2 | | entropy_loss | -2.76 | | explained_variance | 0.961 | | learning_rate | 0.0003 | | loss | -0.0331 | | n_updates | 610 | | policy_gradient_loss | -0.0023 | | std | 0.966 | | value_loss | 0.00282 | ---------------------------------------- [Diag @ 1,000,000 | n_sheep=2 | success=0%] COMPACT_CANT_DRIVE 14/20 NEVER_COMPACT 6/20 action_mag mean=0.216 p10=0.000 p90=0.805 (0=stopped, 1=full speed) min_flock_radius mean=3.39m best=0.00m (target <5m to compact) min_dog_to_com mean=1.18m best=0.11m (FLEE_DIST=7m) min_com_to_pen mean=13.11m best=7.44m reward/step (mean): progress=-0.0011 alignment=+0.0106 pen_bonus=+0.0003 step_cost=-0.0200 complete=+0.0000 -------------------------------- | time/ | | | fps | 2057 | | iterations | 62 | | time_elapsed | 493 | | total_timesteps | 1015808 | -------------------------------- --------------------------------------- | time/ | | | fps | 2067 | | iterations | 63 | | time_elapsed | 499 | | total_timesteps | 1032192 | | train/ | | | approx_kl | 0.008683 | | clip_fraction | 0.0967 | | clip_range | 0.2 | | entropy_loss | -2.77 | | explained_variance | 0.93 | | learning_rate | 0.0003 | | loss | -0.029 | | n_updates | 620 | | policy_gradient_loss | -0.000765 | | std | 0.965 | | value_loss | 0.00446 | --------------------------------------- ----------------------------------------- | time/ | | | fps | 2077 | | iterations | 64 | | time_elapsed | 504 | | total_timesteps | 1048576 | | train/ | | | approx_kl | 0.009014329 | | clip_fraction | 0.113 | | clip_range | 0.2 | | entropy_loss | -2.76 | | explained_variance | 0.984 | | learning_rate | 0.0003 | | loss | -0.0279 | | n_updates | 630 | | policy_gradient_loss | -0.00211 | | std | 0.962 | | value_loss | 0.00312 | ----------------------------------------- Eval num_timesteps=1050000, episode_reward=-31.51 +/- 42.52 Episode length: 2000.00 +/- 0.00 ----------------------------------------- | eval/ | | | mean_ep_length | 2e+03 | | mean_reward | -31.5 | | time/ | | | total_timesteps | 1050000 | | train/ | | | approx_kl | 0.008500135 | | clip_fraction | 0.105 | | clip_range | 0.2 | | entropy_loss | -2.75 | | explained_variance | 0.968 | | learning_rate | 0.0003 | | loss | -0.0306 | | n_updates | 640 | | policy_gradient_loss | -0.00312 | | std | 0.955 | | value_loss | 0.00288 | ----------------------------------------- -------------------------------- | time/ | | | fps | 2042 | | iterations | 65 | | time_elapsed | 521 | | total_timesteps | 1064960 | -------------------------------- ------------------------------------------ | time/ | | | fps | 2056 | | iterations | 66 | | time_elapsed | 525 | | total_timesteps | 1081344 | | train/ | | | approx_kl | 0.0069593494 | | clip_fraction | 0.0923 | | clip_range | 0.2 | | entropy_loss | -2.74 | | explained_variance | 0.835 | | learning_rate | 0.0003 | | loss | -0.0291 | | n_updates | 650 | | policy_gradient_loss | -0.000469 | | std | 0.952 | | value_loss | 0.00186 | ------------------------------------------ ----------------------------------------- | time/ | | | fps | 2064 | | iterations | 67 | | time_elapsed | 531 | | total_timesteps | 1097728 | | train/ | | | approx_kl | 0.007817726 | | clip_fraction | 0.0933 | | clip_range | 0.2 | | entropy_loss | -2.74 | | explained_variance | 0.922 | | learning_rate | 0.0003 | | loss | -0.0206 | | n_updates | 660 | | policy_gradient_loss | -0.00208 | | std | 0.953 | | value_loss | 0.00234 | ----------------------------------------- Eval num_timesteps=1100000, episode_reward=-22.82 +/- 33.61 Episode length: 2000.00 +/- 0.00 ----------------------------------------- | eval/ | | | mean_ep_length | 2e+03 | | mean_reward | -22.8 | | time/ | | | total_timesteps | 1100000 | | train/ | | | approx_kl | 0.006177975 | | clip_fraction | 0.0806 | | clip_range | 0.2 | | entropy_loss | -2.74 | | explained_variance | 0.951 | | learning_rate | 0.0003 | | loss | -0.026 | | n_updates | 670 | | policy_gradient_loss | -5.8e-05 | | std | 0.951 | | value_loss | 0.00184 | ----------------------------------------- -------------------------------- | time/ | | | fps | 2035 | | iterations | 68 | | time_elapsed | 547 | | total_timesteps | 1114112 | -------------------------------- ----------------------------------------- | time/ | | | fps | 2048 | | iterations | 69 | | time_elapsed | 551 | | total_timesteps | 1130496 | | train/ | | | approx_kl | 0.009605391 | | clip_fraction | 0.102 | | clip_range | 0.2 | | entropy_loss | -2.74 | | explained_variance | 0.951 | | learning_rate | 0.0003 | | loss | -0.0344 | | n_updates | 680 | | policy_gradient_loss | -0.0022 | | std | 0.957 | | value_loss | 0.00221 | ----------------------------------------- ------------------------------------------ | time/ | | | fps | 2060 | | iterations | 70 | | time_elapsed | 556 | | total_timesteps | 1146880 | | train/ | | | approx_kl | 0.0064521013 | | clip_fraction | 0.0953 | | clip_range | 0.2 | | entropy_loss | -2.75 | | explained_variance | 0.898 | | learning_rate | 0.0003 | | loss | -0.0348 | | n_updates | 690 | | policy_gradient_loss | -0.00112 | | std | 0.96 | | value_loss | 0.00221 | ------------------------------------------ Eval num_timesteps=1150000, episode_reward=-26.36 +/- 35.49 Episode length: 2000.00 +/- 0.00 ---------------------------------------- | eval/ | | | mean_ep_length | 2e+03 | | mean_reward | -26.4 | | time/ | | | total_timesteps | 1150000 | | train/ | | | approx_kl | 0.00777065 | | clip_fraction | 0.0837 | | clip_range | 0.2 | | entropy_loss | -2.76 | | explained_variance | 0.907 | | learning_rate | 0.0003 | | loss | -0.0198 | | n_updates | 700 | | policy_gradient_loss | -0.000371 | | std | 0.963 | | value_loss | 0.00182 | ---------------------------------------- -------------------------------- | time/ | | | fps | 2031 | | iterations | 71 | | time_elapsed | 572 | | total_timesteps | 1163264 | -------------------------------- --------------------------------------- | time/ | | | fps | 2044 | | iterations | 72 | | time_elapsed | 577 | | total_timesteps | 1179648 | | train/ | | | approx_kl | 0.006344 | | clip_fraction | 0.0719 | | clip_range | 0.2 | | entropy_loss | -2.76 | | explained_variance | 0.908 | | learning_rate | 0.0003 | | loss | -0.0347 | | n_updates | 710 | | policy_gradient_loss | -0.000455 | | std | 0.961 | | value_loss | 0.00145 | --------------------------------------- ------------------------------------------ | time/ | | | fps | 2054 | | iterations | 73 | | time_elapsed | 582 | | total_timesteps | 1196032 | | train/ | | | approx_kl | 0.0060829036 | | clip_fraction | 0.0854 | | clip_range | 0.2 | | entropy_loss | -2.75 | | explained_variance | 0.896 | | learning_rate | 0.0003 | | loss | -0.0232 | | n_updates | 720 | | policy_gradient_loss | -0.00108 | | std | 0.957 | | value_loss | 0.00152 | ------------------------------------------ Eval num_timesteps=1200000, episode_reward=-14.33 +/- 30.83 Episode length: 2000.00 +/- 0.00 ------------------------------------------ | eval/ | | | mean_ep_length | 2e+03 | | mean_reward | -14.3 | | time/ | | | total_timesteps | 1200000 | | train/ | | | approx_kl | 0.0073732347 | | clip_fraction | 0.0783 | | clip_range | 0.2 | | entropy_loss | -2.76 | | explained_variance | 0.948 | | learning_rate | 0.0003 | | loss | -0.0267 | | n_updates | 730 | | policy_gradient_loss | -0.00212 | | std | 0.968 | | value_loss | 0.00253 | ------------------------------------------ [Diag @ 1,200,000 | n_sheep=2 | success=0%] COMPACT_CANT_DRIVE 15/20 NEVER_COMPACT 5/20 action_mag mean=0.273 p10=0.004 p90=1.008 (0=stopped, 1=full speed) min_flock_radius mean=3.94m best=0.97m (target <5m to compact) min_dog_to_com mean=1.16m best=0.35m (FLEE_DIST=7m) min_com_to_pen mean=13.54m best=4.20m reward/step (mean): progress=+0.0001 alignment=+0.0121 pen_bonus=+0.0000 step_cost=-0.0200 complete=+0.0000 -------------------------------- | time/ | | | fps | 1998 | | iterations | 74 | | time_elapsed | 606 | | total_timesteps | 1212416 | -------------------------------- ----------------------------------------- | time/ | | | fps | 2008 | | iterations | 75 | | time_elapsed | 611 | | total_timesteps | 1228800 | | train/ | | | approx_kl | 0.006109112 | | clip_fraction | 0.0814 | | clip_range | 0.2 | | entropy_loss | -2.78 | | explained_variance | 0.86 | | learning_rate | 0.0003 | | loss | -0.0205 | | n_updates | 740 | | policy_gradient_loss | -0.000541 | | std | 0.973 | | value_loss | 0.00171 | ----------------------------------------- ---------------------------------------- | time/ | | | fps | 2016 | | iterations | 76 | | time_elapsed | 617 | | total_timesteps | 1245184 | | train/ | | | approx_kl | 0.00703271 | | clip_fraction | 0.0781 | | clip_range | 0.2 | | entropy_loss | -2.78 | | explained_variance | 0.934 | | learning_rate | 0.0003 | | loss | -0.0394 | | n_updates | 750 | | policy_gradient_loss | -0.00105 | | std | 0.975 | | value_loss | 0.00168 | ---------------------------------------- Eval num_timesteps=1250000, episode_reward=-18.12 +/- 39.82 Episode length: 2000.00 +/- 0.00 ------------------------------------------ | eval/ | | | mean_ep_length | 2e+03 | | mean_reward | -18.1 | | time/ | | | total_timesteps | 1250000 | | train/ | | | approx_kl | 0.0064994176 | | clip_fraction | 0.0698 | | clip_range | 0.2 | | entropy_loss | -2.8 | | explained_variance | 0.919 | | learning_rate | 0.0003 | | loss | -0.0166 | | n_updates | 760 | | policy_gradient_loss | -0.000919 | | std | 0.985 | | value_loss | 0.000832 | ------------------------------------------ -------------------------------- | time/ | | | fps | 1989 | | iterations | 77 | | time_elapsed | 634 | | total_timesteps | 1261568 | -------------------------------- ----------------------------------------- | time/ | | | fps | 2001 | | iterations | 78 | | time_elapsed | 638 | | total_timesteps | 1277952 | | train/ | | | approx_kl | 0.008321709 | | clip_fraction | 0.0902 | | clip_range | 0.2 | | entropy_loss | -2.81 | | explained_variance | 0.874 | | learning_rate | 0.0003 | | loss | -0.0295 | | n_updates | 770 | | policy_gradient_loss | -0.00219 | | std | 0.991 | | value_loss | 0.00127 | ----------------------------------------- ----------------------------------------- | time/ | | | fps | 2010 | | iterations | 79 | | time_elapsed | 643 | | total_timesteps | 1294336 | | train/ | | | approx_kl | 0.009220061 | | clip_fraction | 0.112 | | clip_range | 0.2 | | entropy_loss | -2.82 | | explained_variance | 0.952 | | learning_rate | 0.0003 | | loss | -0.0379 | | n_updates | 780 | | policy_gradient_loss | -0.00411 | | std | 0.994 | | value_loss | 0.00295 | ----------------------------------------- Eval num_timesteps=1300000, episode_reward=-22.41 +/- 35.57 Episode length: 2000.00 +/- 0.00 ------------------------------------------ | eval/ | | | mean_ep_length | 2e+03 | | mean_reward | -22.4 | | time/ | | | total_timesteps | 1300000 | | train/ | | | approx_kl | 0.0071307076 | | clip_fraction | 0.0826 | | clip_range | 0.2 | | entropy_loss | -2.83 | | explained_variance | 0.948 | | learning_rate | 0.0003 | | loss | -0.0281 | | n_updates | 790 | | policy_gradient_loss | -0.00178 | | std | 0.995 | | value_loss | 0.00169 | ------------------------------------------ -------------------------------- | time/ | | | fps | 1986 | | iterations | 80 | | time_elapsed | 659 | | total_timesteps | 1310720 | -------------------------------- ----------------------------------------- | time/ | | | fps | 1996 | | iterations | 81 | | time_elapsed | 664 | | total_timesteps | 1327104 | | train/ | | | approx_kl | 0.008566003 | | clip_fraction | 0.0857 | | clip_range | 0.2 | | entropy_loss | -2.84 | | explained_variance | 0.904 | | learning_rate | 0.0003 | | loss | -0.0369 | | n_updates | 800 | | policy_gradient_loss | -0.00199 | | std | 1.01 | | value_loss | 0.00203 | ----------------------------------------- ------------------------------------------ | time/ | | | fps | 2006 | | iterations | 82 | | time_elapsed | 669 | | total_timesteps | 1343488 | | train/ | | | approx_kl | 0.0082352655 | | clip_fraction | 0.0989 | | clip_range | 0.2 | | entropy_loss | -2.86 | | explained_variance | 0.918 | | learning_rate | 0.0003 | | loss | -0.0297 | | n_updates | 810 | | policy_gradient_loss | -0.0023 | | std | 1.01 | | value_loss | 0.00203 | ------------------------------------------ Eval num_timesteps=1350000, episode_reward=-14.21 +/- 38.53 Episode length: 2000.00 +/- 0.00 ------------------------------------------ | eval/ | | | mean_ep_length | 2e+03 | | mean_reward | -14.2 | | time/ | | | total_timesteps | 1350000 | | train/ | | | approx_kl | 0.0066830693 | | clip_fraction | 0.0831 | | clip_range | 0.2 | | entropy_loss | -2.86 | | explained_variance | 0.923 | | learning_rate | 0.0003 | | loss | -0.0331 | | n_updates | 820 | | policy_gradient_loss | -0.00226 | | std | 1.01 | | value_loss | 0.00125 | ------------------------------------------ -------------------------------- | time/ | | | fps | 1983 | | iterations | 83 | | time_elapsed | 685 | | total_timesteps | 1359872 | -------------------------------- ----------------------------------------- | time/ | | | fps | 1991 | | iterations | 84 | | time_elapsed | 691 | | total_timesteps | 1376256 | | train/ | | | approx_kl | 0.008341949 | | clip_fraction | 0.101 | | clip_range | 0.2 | | entropy_loss | -2.85 | | explained_variance | 0.928 | | learning_rate | 0.0003 | | loss | -0.0156 | | n_updates | 830 | | policy_gradient_loss | -0.00132 | | std | 1.01 | | value_loss | 0.00407 | ----------------------------------------- ----------------------------------------- | time/ | | | fps | 1999 | | iterations | 85 | | time_elapsed | 696 | | total_timesteps | 1392640 | | train/ | | | approx_kl | 0.010089031 | | clip_fraction | 0.109 | | clip_range | 0.2 | | entropy_loss | -2.84 | | explained_variance | 0.914 | | learning_rate | 0.0003 | | loss | -0.0249 | | n_updates | 840 | | policy_gradient_loss | -0.00202 | | std | 0.999 | | value_loss | 0.00555 | ----------------------------------------- Eval num_timesteps=1400000, episode_reward=-5.74 +/- 37.76 Episode length: 2000.00 +/- 0.00 ---------------------------------------- | eval/ | | | mean_ep_length | 2e+03 | | mean_reward | -5.74 | | time/ | | | total_timesteps | 1400000 | | train/ | | | approx_kl | 0.00840036 | | clip_fraction | 0.112 | | clip_range | 0.2 | | entropy_loss | -2.84 | | explained_variance | 0.915 | | learning_rate | 0.0003 | | loss | -0.0267 | | n_updates | 850 | | policy_gradient_loss | -0.00422 | | std | 1 | | value_loss | 0.0017 | ---------------------------------------- [Diag @ 1,400,000 | n_sheep=2 | success=0%] COMPACT_CANT_DRIVE 16/20 NEVER_COMPACT 4/20 action_mag mean=0.258 p10=0.000 p90=1.004 (0=stopped, 1=full speed) min_flock_radius mean=3.30m best=0.61m (target <5m to compact) min_dog_to_com mean=0.76m best=0.22m (FLEE_DIST=7m) min_com_to_pen mean=12.16m best=4.08m reward/step (mean): progress=+0.0035 alignment=+0.0165 pen_bonus=+0.0000 step_cost=-0.0200 complete=+0.0000 -------------------------------- | time/ | | | fps | 1954 | | iterations | 86 | | time_elapsed | 720 | | total_timesteps | 1409024 | -------------------------------- ----------------------------------------- | time/ | | | fps | 1964 | | iterations | 87 | | time_elapsed | 725 | | total_timesteps | 1425408 | | train/ | | | approx_kl | 0.007908808 | | clip_fraction | 0.0839 | | clip_range | 0.2 | | entropy_loss | -2.85 | | explained_variance | 0.755 | | learning_rate | 0.0003 | | loss | -0.018 | | n_updates | 860 | | policy_gradient_loss | -0.00223 | | std | 1.01 | | value_loss | 0.00248 | ----------------------------------------- ----------------------------------------- | time/ | | | fps | 1972 | | iterations | 88 | | time_elapsed | 730 | | total_timesteps | 1441792 | | train/ | | | approx_kl | 0.007957449 | | clip_fraction | 0.0864 | | clip_range | 0.2 | | entropy_loss | -2.86 | | explained_variance | 0.868 | | learning_rate | 0.0003 | | loss | -0.0315 | | n_updates | 870 | | policy_gradient_loss | -0.00288 | | std | 1.01 | | value_loss | 0.00145 | ----------------------------------------- Eval num_timesteps=1450000, episode_reward=-13.10 +/- 29.51 Episode length: 2000.00 +/- 0.00 ----------------------------------------- | eval/ | | | mean_ep_length | 2e+03 | | mean_reward | -13.1 | | time/ | | | total_timesteps | 1450000 | | train/ | | | approx_kl | 0.007803983 | | clip_fraction | 0.083 | | clip_range | 0.2 | | entropy_loss | -2.86 | | explained_variance | 0.83 | | learning_rate | 0.0003 | | loss | -0.0212 | | n_updates | 880 | | policy_gradient_loss | -0.00119 | | std | 1.01 | | value_loss | 0.00191 | ----------------------------------------- -------------------------------- | time/ | | | fps | 1952 | | iterations | 89 | | time_elapsed | 746 | | total_timesteps | 1458176 | -------------------------------- ----------------------------------------- | time/ | | | fps | 1961 | | iterations | 90 | | time_elapsed | 751 | | total_timesteps | 1474560 | | train/ | | | approx_kl | 0.010021031 | | clip_fraction | 0.097 | | clip_range | 0.2 | | entropy_loss | -2.88 | | explained_variance | 0.902 | | learning_rate | 0.0003 | | loss | -0.0221 | | n_updates | 890 | | policy_gradient_loss | -0.00294 | | std | 1.02 | | value_loss | 0.00136 | ----------------------------------------- ------------------------------------------ | time/ | | | fps | 1970 | | iterations | 91 | | time_elapsed | 756 | | total_timesteps | 1490944 | | train/ | | | approx_kl | 0.0076614916 | | clip_fraction | 0.0963 | | clip_range | 0.2 | | entropy_loss | -2.89 | | explained_variance | 0.945 | | learning_rate | 0.0003 | | loss | -0.0273 | | n_updates | 900 | | policy_gradient_loss | -0.00355 | | std | 1.03 | | value_loss | 0.00181 | ------------------------------------------ Eval num_timesteps=1500000, episode_reward=5.01 +/- 34.23 Episode length: 2000.00 +/- 0.00 ----------------------------------------- | eval/ | | | mean_ep_length | 2e+03 | | mean_reward | 5.01 | | time/ | | | total_timesteps | 1500000 | | train/ | | | approx_kl | 0.005815446 | | clip_fraction | 0.0675 | | clip_range | 0.2 | | entropy_loss | -2.9 | | explained_variance | 0.934 | | learning_rate | 0.0003 | | loss | -0.0316 | | n_updates | 910 | | policy_gradient_loss | -0.00215 | | std | 1.03 | | value_loss | 0.00162 | ----------------------------------------- -------------------------------- | time/ | | | fps | 1950 | | iterations | 92 | | time_elapsed | 772 | | total_timesteps | 1507328 | -------------------------------- ------------------------------------------ | time/ | | | fps | 1959 | | iterations | 93 | | time_elapsed | 777 | | total_timesteps | 1523712 | | train/ | | | approx_kl | 0.0071218535 | | clip_fraction | 0.0897 | | clip_range | 0.2 | | entropy_loss | -2.9 | | explained_variance | 0.937 | | learning_rate | 0.0003 | | loss | -0.0219 | | n_updates | 920 | | policy_gradient_loss | -0.00225 | | std | 1.03 | | value_loss | 0.00463 | ------------------------------------------ ----------------------------------------- | time/ | | | fps | 1967 | | iterations | 94 | | time_elapsed | 782 | | total_timesteps | 1540096 | | train/ | | | approx_kl | 0.006857206 | | clip_fraction | 0.0809 | | clip_range | 0.2 | | entropy_loss | -2.89 | | explained_variance | 0.933 | | learning_rate | 0.0003 | | loss | -0.0252 | | n_updates | 930 | | policy_gradient_loss | -0.00219 | | std | 1.02 | | value_loss | 0.00436 | ----------------------------------------- Eval num_timesteps=1550000, episode_reward=-4.04 +/- 33.69 Episode length: 2000.00 +/- 0.00 ----------------------------------------- | eval/ | | | mean_ep_length | 2e+03 | | mean_reward | -4.04 | | time/ | | | total_timesteps | 1550000 | | train/ | | | approx_kl | 0.006146897 | | clip_fraction | 0.0821 | | clip_range | 0.2 | | entropy_loss | -2.87 | | explained_variance | 0.913 | | learning_rate | 0.0003 | | loss | -0.0352 | | n_updates | 940 | | policy_gradient_loss | -0.00258 | | std | 1.02 | | value_loss | 0.00325 | ----------------------------------------- -------------------------------- | time/ | | | fps | 1948 | | iterations | 95 | | time_elapsed | 798 | | total_timesteps | 1556480 | -------------------------------- ------------------------------------------ | time/ | | | fps | 1958 | | iterations | 96 | | time_elapsed | 803 | | total_timesteps | 1572864 | | train/ | | | approx_kl | 0.0069321445 | | clip_fraction | 0.0778 | | clip_range | 0.2 | | entropy_loss | -2.86 | | explained_variance | 0.94 | | learning_rate | 0.0003 | | loss | -0.013 | | n_updates | 950 | | policy_gradient_loss | -0.00214 | | std | 1.01 | | value_loss | 0.00162 | ------------------------------------------ ------------------------------------------ | time/ | | | fps | 1965 | | iterations | 97 | | time_elapsed | 808 | | total_timesteps | 1589248 | | train/ | | | approx_kl | 0.0066491435 | | clip_fraction | 0.0714 | | clip_range | 0.2 | | entropy_loss | -2.88 | | explained_variance | 0.941 | | learning_rate | 0.0003 | | loss | -0.0304 | | n_updates | 960 | | policy_gradient_loss | -0.00212 | | std | 1.03 | | value_loss | 0.0011 | ------------------------------------------ Eval num_timesteps=1600000, episode_reward=12.65 +/- 31.73 Episode length: 2000.00 +/- 0.00 ------------------------------------------ | eval/ | | | mean_ep_length | 2e+03 | | mean_reward | 12.6 | | time/ | | | total_timesteps | 1600000 | | train/ | | | approx_kl | 0.0050257677 | | clip_fraction | 0.0588 | | clip_range | 0.2 | | entropy_loss | -2.9 | | explained_variance | 0.939 | | learning_rate | 0.0003 | | loss | -0.0359 | | n_updates | 970 | | policy_gradient_loss | -0.0013 | | std | 1.04 | | value_loss | 0.00201 | ------------------------------------------ [Diag @ 1,600,000 | n_sheep=2 | success=0%] COMPACT_CANT_DRIVE 13/20 NEVER_COMPACT 7/20 action_mag mean=0.252 p10=0.004 p90=0.980 (0=stopped, 1=full speed) min_flock_radius mean=4.30m best=0.92m (target <5m to compact) min_dog_to_com mean=0.74m best=0.38m (FLEE_DIST=7m) min_com_to_pen mean=13.76m best=5.49m reward/step (mean): progress=-0.0006 alignment=+0.0287 pen_bonus=+0.0000 step_cost=-0.0200 complete=+0.0000 [Curriculum] leaving stage n_sheep=2 after 800,000 steps | training success rate (last 100 eps) = 0% [Curriculum] → 3 sheep at step 1,600,000 -------------------------------- | time/ | | | fps | 1930 | | iterations | 98 | | time_elapsed | 831 | | total_timesteps | 1605632 | -------------------------------- ------------------------------------------ | time/ | | | fps | 1937 | | iterations | 99 | | time_elapsed | 837 | | total_timesteps | 1622016 | | train/ | | | approx_kl | 0.0085028205 | | clip_fraction | 0.0905 | | clip_range | 0.2 | | entropy_loss | -2.89 | | explained_variance | 0.909 | | learning_rate | 0.0003 | | loss | -0.0346 | | n_updates | 980 | | policy_gradient_loss | -0.00245 | | std | 1.02 | | value_loss | 0.00492 | ------------------------------------------ ----------------------------------------- | time/ | | | fps | 1945 | | iterations | 100 | | time_elapsed | 842 | | total_timesteps | 1638400 | | train/ | | | approx_kl | 0.009084044 | | clip_fraction | 0.118 | | clip_range | 0.2 | | entropy_loss | -2.91 | | explained_variance | 0.964 | | learning_rate | 0.0003 | | loss | -0.0416 | | n_updates | 990 | | policy_gradient_loss | 0.0025 | | std | 1.04 | | value_loss | 0.00194 | ----------------------------------------- Eval num_timesteps=1650000, episode_reward=3.05 +/- 36.42 Episode length: 2000.00 +/- 0.00 ----------------------------------------- | eval/ | | | mean_ep_length | 2e+03 | | mean_reward | 3.05 | | time/ | | | total_timesteps | 1650000 | | train/ | | | approx_kl | 0.009275759 | | clip_fraction | 0.108 | | clip_range | 0.2 | | entropy_loss | -2.92 | | explained_variance | 0.965 | | learning_rate | 0.0003 | | loss | -0.0336 | | n_updates | 1000 | | policy_gradient_loss | 0.000149 | | std | 1.04 | | value_loss | 0.00185 | ----------------------------------------- -------------------------------- | time/ | | | fps | 1926 | | iterations | 101 | | time_elapsed | 859 | | total_timesteps | 1654784 | -------------------------------- ----------------------------------------- | time/ | | | fps | 1934 | | iterations | 102 | | time_elapsed | 864 | | total_timesteps | 1671168 | | train/ | | | approx_kl | 0.008650862 | | clip_fraction | 0.117 | | clip_range | 0.2 | | entropy_loss | -2.92 | | explained_variance | 0.938 | | learning_rate | 0.0003 | | loss | -0.0279 | | n_updates | 1010 | | policy_gradient_loss | -0.000545 | | std | 1.04 | | value_loss | 0.00611 | ----------------------------------------- --------------------------------------- | time/ | | | fps | 1939 | | iterations | 103 | | time_elapsed | 869 | | total_timesteps | 1687552 | | train/ | | | approx_kl | 0.0080826 | | clip_fraction | 0.0992 | | clip_range | 0.2 | | entropy_loss | -2.93 | | explained_variance | 0.952 | | learning_rate | 0.0003 | | loss | -0.0415 | | n_updates | 1020 | | policy_gradient_loss | -0.00201 | | std | 1.05 | | value_loss | 0.00251 | --------------------------------------- Eval num_timesteps=1700000, episode_reward=-4.66 +/- 36.05 Episode length: 2000.00 +/- 0.00 ---------------------------------------- | eval/ | | | mean_ep_length | 2e+03 | | mean_reward | -4.66 | | time/ | | | total_timesteps | 1700000 | | train/ | | | approx_kl | 0.00786162 | | clip_fraction | 0.0921 | | clip_range | 0.2 | | entropy_loss | -2.95 | | explained_variance | 0.893 | | learning_rate | 0.0003 | | loss | -0.0301 | | n_updates | 1030 | | policy_gradient_loss | -0.000631 | | std | 1.06 | | value_loss | 0.00158 | ---------------------------------------- -------------------------------- | time/ | | | fps | 1922 | | iterations | 104 | | time_elapsed | 886 | | total_timesteps | 1703936 | -------------------------------- ----------------------------------------- | time/ | | | fps | 1930 | | iterations | 105 | | time_elapsed | 891 | | total_timesteps | 1720320 | | train/ | | | approx_kl | 0.008055547 | | clip_fraction | 0.0842 | | clip_range | 0.2 | | entropy_loss | -2.96 | | explained_variance | 0.918 | | learning_rate | 0.0003 | | loss | -0.027 | | n_updates | 1040 | | policy_gradient_loss | -6.56e-05 | | std | 1.07 | | value_loss | 0.00193 | ----------------------------------------- ----------------------------------------- | time/ | | | fps | 1937 | | iterations | 106 | | time_elapsed | 896 | | total_timesteps | 1736704 | | train/ | | | approx_kl | 0.008067045 | | clip_fraction | 0.087 | | clip_range | 0.2 | | entropy_loss | -2.97 | | explained_variance | 0.878 | | learning_rate | 0.0003 | | loss | -0.0281 | | n_updates | 1050 | | policy_gradient_loss | -0.00194 | | std | 1.07 | | value_loss | 0.0082 | ----------------------------------------- Eval num_timesteps=1750000, episode_reward=-0.31 +/- 42.66 Episode length: 2000.00 +/- 0.00 ------------------------------------------ | eval/ | | | mean_ep_length | 2e+03 | | mean_reward | -0.309 | | time/ | | | total_timesteps | 1750000 | | train/ | | | approx_kl | 0.0066514863 | | clip_fraction | 0.0808 | | clip_range | 0.2 | | entropy_loss | -2.99 | | explained_variance | 0.888 | | learning_rate | 0.0003 | | loss | -0.0335 | | n_updates | 1060 | | policy_gradient_loss | -0.00108 | | std | 1.08 | | value_loss | 0.00303 | ------------------------------------------ -------------------------------- | time/ | | | fps | 1921 | | iterations | 107 | | time_elapsed | 912 | | total_timesteps | 1753088 | -------------------------------- ----------------------------------------- | time/ | | | fps | 1927 | | iterations | 108 | | time_elapsed | 917 | | total_timesteps | 1769472 | | train/ | | | approx_kl | 0.008252729 | | clip_fraction | 0.093 | | clip_range | 0.2 | | entropy_loss | -3 | | explained_variance | 0.959 | | learning_rate | 0.0003 | | loss | -0.0413 | | n_updates | 1070 | | policy_gradient_loss | -0.00241 | | std | 1.09 | | value_loss | 0.00122 | ----------------------------------------- ------------------------------------------ | time/ | | | fps | 1935 | | iterations | 109 | | time_elapsed | 922 | | total_timesteps | 1785856 | | train/ | | | approx_kl | 0.0073527684 | | clip_fraction | 0.0822 | | clip_range | 0.2 | | entropy_loss | -3.01 | | explained_variance | 0.883 | | learning_rate | 0.0003 | | loss | -0.018 | | n_updates | 1080 | | policy_gradient_loss | -0.00172 | | std | 1.1 | | value_loss | 0.00172 | ------------------------------------------ Eval num_timesteps=1800000, episode_reward=8.99 +/- 39.35 Episode length: 2000.00 +/- 0.00 ----------------------------------------- | eval/ | | | mean_ep_length | 2e+03 | | mean_reward | 8.99 | | time/ | | | total_timesteps | 1800000 | | train/ | | | approx_kl | 0.006149094 | | clip_fraction | 0.0771 | | clip_range | 0.2 | | entropy_loss | -3.03 | | explained_variance | 0.911 | | learning_rate | 0.0003 | | loss | -0.0315 | | n_updates | 1090 | | policy_gradient_loss | -0.000744 | | std | 1.1 | | value_loss | 0.00456 | ----------------------------------------- [Diag @ 1,800,000 | n_sheep=3 | success=0%] NEVER_COMPACT 19/20 COMPACT_CANT_DRIVE 1/20 action_mag mean=0.049 p10=0.007 p90=0.049 (0=stopped, 1=full speed) min_flock_radius mean=7.79m best=4.73m (target <5m to compact) min_dog_to_com mean=0.92m best=0.25m (FLEE_DIST=7m) min_com_to_pen mean=14.27m best=7.54m reward/step (mean): progress=-0.0043 alignment=+0.0208 pen_bonus=+0.0000 step_cost=-0.0200 complete=+0.0000 -------------------------------- | time/ | | | fps | 1899 | | iterations | 110 | | time_elapsed | 948 | | total_timesteps | 1802240 | -------------------------------- ----------------------------------------- | time/ | | | fps | 1906 | | iterations | 111 | | time_elapsed | 953 | | total_timesteps | 1818624 | | train/ | | | approx_kl | 0.007161974 | | clip_fraction | 0.0871 | | clip_range | 0.2 | | entropy_loss | -3.03 | | explained_variance | 0.914 | | learning_rate | 0.0003 | | loss | -0.0359 | | n_updates | 1100 | | policy_gradient_loss | -0.00186 | | std | 1.1 | | value_loss | 0.00214 | ----------------------------------------- ---------------------------------------- | time/ | | | fps | 1914 | | iterations | 112 | | time_elapsed | 958 | | total_timesteps | 1835008 | | train/ | | | approx_kl | 0.00886854 | | clip_fraction | 0.103 | | clip_range | 0.2 | | entropy_loss | -3.04 | | explained_variance | 0.94 | | learning_rate | 0.0003 | | loss | -0.04 | | n_updates | 1110 | | policy_gradient_loss | -0.00333 | | std | 1.11 | | value_loss | 0.00456 | ---------------------------------------- Eval num_timesteps=1850000, episode_reward=14.49 +/- 36.35 Episode length: 2000.00 +/- 0.00 ------------------------------------------ | eval/ | | | mean_ep_length | 2e+03 | | mean_reward | 14.5 | | time/ | | | total_timesteps | 1850000 | | train/ | | | approx_kl | 0.0058414284 | | clip_fraction | 0.0642 | | clip_range | 0.2 | | entropy_loss | -3.05 | | explained_variance | 0.871 | | learning_rate | 0.0003 | | loss | -0.033 | | n_updates | 1120 | | policy_gradient_loss | -0.000891 | | std | 1.11 | | value_loss | 0.00394 | ------------------------------------------ -------------------------------- | time/ | | | fps | 1898 | | iterations | 113 | | time_elapsed | 975 | | total_timesteps | 1851392 | -------------------------------- ----------------------------------------- | time/ | | | fps | 1906 | | iterations | 114 | | time_elapsed | 979 | | total_timesteps | 1867776 | | train/ | | | approx_kl | 0.008916938 | | clip_fraction | 0.0916 | | clip_range | 0.2 | | entropy_loss | -3.05 | | explained_variance | 0.937 | | learning_rate | 0.0003 | | loss | -0.0334 | | n_updates | 1130 | | policy_gradient_loss | -0.00257 | | std | 1.12 | | value_loss | 0.00285 | ----------------------------------------- ----------------------------------------- | time/ | | | fps | 1913 | | iterations | 115 | | time_elapsed | 984 | | total_timesteps | 1884160 | | train/ | | | approx_kl | 0.008523149 | | clip_fraction | 0.0907 | | clip_range | 0.2 | | entropy_loss | -3.06 | | explained_variance | 0.954 | | learning_rate | 0.0003 | | loss | -0.0339 | | n_updates | 1140 | | policy_gradient_loss | -0.0034 | | std | 1.12 | | value_loss | 0.00209 | ----------------------------------------- Eval num_timesteps=1900000, episode_reward=9.85 +/- 42.18 Episode length: 2000.00 +/- 0.00 ------------------------------------------ | eval/ | | | mean_ep_length | 2e+03 | | mean_reward | 9.85 | | time/ | | | total_timesteps | 1900000 | | train/ | | | approx_kl | 0.0075978916 | | clip_fraction | 0.0819 | | clip_range | 0.2 | | entropy_loss | -3.06 | | explained_variance | 0.96 | | learning_rate | 0.0003 | | loss | -0.0313 | | n_updates | 1150 | | policy_gradient_loss | -0.00272 | | std | 1.12 | | value_loss | 0.00332 | ------------------------------------------ -------------------------------- | time/ | | | fps | 1896 | | iterations | 116 | | time_elapsed | 1002 | | total_timesteps | 1900544 | -------------------------------- ----------------------------------------- | time/ | | | fps | 1902 | | iterations | 117 | | time_elapsed | 1007 | | total_timesteps | 1916928 | | train/ | | | approx_kl | 0.008376695 | | clip_fraction | 0.0935 | | clip_range | 0.2 | | entropy_loss | -3.07 | | explained_variance | 0.964 | | learning_rate | 0.0003 | | loss | -0.0392 | | n_updates | 1160 | | policy_gradient_loss | -0.00354 | | std | 1.12 | | value_loss | 0.00203 | ----------------------------------------- ------------------------------------------ | time/ | | | fps | 1909 | | iterations | 118 | | time_elapsed | 1012 | | total_timesteps | 1933312 | | train/ | | | approx_kl | 0.0077100536 | | clip_fraction | 0.0854 | | clip_range | 0.2 | | entropy_loss | -3.07 | | explained_variance | 0.933 | | learning_rate | 0.0003 | | loss | -0.0467 | | n_updates | 1170 | | policy_gradient_loss | -0.00421 | | std | 1.12 | | value_loss | 0.00132 | ------------------------------------------ ----------------------------------------- | time/ | | | fps | 1915 | | iterations | 119 | | time_elapsed | 1018 | | total_timesteps | 1949696 | | train/ | | | approx_kl | 0.006848542 | | clip_fraction | 0.0674 | | clip_range | 0.2 | | entropy_loss | -3.07 | | explained_variance | 0.959 | | learning_rate | 0.0003 | | loss | -0.0335 | | n_updates | 1180 | | policy_gradient_loss | -0.00229 | | std | 1.13 | | value_loss | 0.00138 | ----------------------------------------- Eval num_timesteps=1950000, episode_reward=29.72 +/- 38.42 Episode length: 2000.00 +/- 0.00 ----------------------------------------- | eval/ | | | mean_ep_length | 2e+03 | | mean_reward | 29.7 | | time/ | | | total_timesteps | 1950000 | | train/ | | | approx_kl | 0.007300608 | | clip_fraction | 0.0824 | | clip_range | 0.2 | | entropy_loss | -3.07 | | explained_variance | 0.977 | | learning_rate | 0.0003 | | loss | -0.0358 | | n_updates | 1190 | | policy_gradient_loss | -0.00364 | | std | 1.12 | | value_loss | 0.00159 | ----------------------------------------- -------------------------------- | time/ | | | fps | 1899 | | iterations | 120 | | time_elapsed | 1034 | | total_timesteps | 1966080 | -------------------------------- ------------------------------------------ | time/ | | | fps | 1906 | | iterations | 121 | | time_elapsed | 1040 | | total_timesteps | 1982464 | | train/ | | | approx_kl | 0.0072772675 | | clip_fraction | 0.0703 | | clip_range | 0.2 | | entropy_loss | -3.07 | | explained_variance | 0.882 | | learning_rate | 0.0003 | | loss | -0.0357 | | n_updates | 1200 | | policy_gradient_loss | -0.00163 | | std | 1.13 | | value_loss | 0.00471 | ------------------------------------------ ----------------------------------------- | time/ | | | fps | 1912 | | iterations | 122 | | time_elapsed | 1045 | | total_timesteps | 1998848 | | train/ | | | approx_kl | 0.007866079 | | clip_fraction | 0.0898 | | clip_range | 0.2 | | entropy_loss | -3.07 | | explained_variance | 0.962 | | learning_rate | 0.0003 | | loss | -0.0304 | | n_updates | 1210 | | policy_gradient_loss | -0.0052 | | std | 1.13 | | value_loss | 0.0014 | ----------------------------------------- Eval num_timesteps=2000000, episode_reward=14.20 +/- 34.02 Episode length: 2000.00 +/- 0.00 ------------------------------------------ | eval/ | | | mean_ep_length | 2e+03 | | mean_reward | 14.2 | | time/ | | | total_timesteps | 2000000 | | train/ | | | approx_kl | 0.0073383995 | | clip_fraction | 0.083 | | clip_range | 0.2 | | entropy_loss | -3.07 | | explained_variance | 0.95 | | learning_rate | 0.0003 | | loss | -0.0369 | | n_updates | 1220 | | policy_gradient_loss | -0.00296 | | std | 1.12 | | value_loss | 0.00336 | ------------------------------------------ [Diag @ 2,000,000 | n_sheep=3 | success=0%] NEVER_COMPACT 12/20 COMPACT_CANT_DRIVE 8/20 action_mag mean=0.076 p10=0.007 p90=0.097 (0=stopped, 1=full speed) min_flock_radius mean=5.33m best=0.00m (target <5m to compact) min_dog_to_com mean=1.01m best=0.16m (FLEE_DIST=7m) min_com_to_pen mean=12.40m best=6.50m reward/step (mean): progress=+0.0041 alignment=+0.0263 pen_bonus=+0.0013 step_cost=-0.0200 complete=+0.0000 -------------------------------- | time/ | | | fps | 1881 | | iterations | 123 | | time_elapsed | 1071 | | total_timesteps | 2015232 | -------------------------------- ------------------------------------------ | time/ | | | fps | 1887 | | iterations | 124 | | time_elapsed | 1076 | | total_timesteps | 2031616 | | train/ | | | approx_kl | 0.0060287267 | | clip_fraction | 0.0716 | | clip_range | 0.2 | | entropy_loss | -3.07 | | explained_variance | 0.902 | | learning_rate | 0.0003 | | loss | -0.0402 | | n_updates | 1230 | | policy_gradient_loss | -0.00308 | | std | 1.13 | | value_loss | 0.00475 | ------------------------------------------ ------------------------------------------ | time/ | | | fps | 1894 | | iterations | 125 | | time_elapsed | 1081 | | total_timesteps | 2048000 | | train/ | | | approx_kl | 0.0073304214 | | clip_fraction | 0.08 | | clip_range | 0.2 | | entropy_loss | -3.08 | | explained_variance | 0.95 | | learning_rate | 0.0003 | | loss | -0.0436 | | n_updates | 1240 | | policy_gradient_loss | -0.00373 | | std | 1.13 | | value_loss | 0.00138 | ------------------------------------------ Eval num_timesteps=2050000, episode_reward=18.68 +/- 36.20 Episode length: 2000.00 +/- 0.00 ------------------------------------------ | eval/ | | | mean_ep_length | 2e+03 | | mean_reward | 18.7 | | time/ | | | total_timesteps | 2050000 | | train/ | | | approx_kl | 0.0068036346 | | clip_fraction | 0.0768 | | clip_range | 0.2 | | entropy_loss | -3.07 | | explained_variance | 0.897 | | learning_rate | 0.0003 | | loss | -0.0461 | | n_updates | 1250 | | policy_gradient_loss | -0.00392 | | std | 1.13 | | value_loss | 0.0013 | ------------------------------------------ -------------------------------- | time/ | | | fps | 1880 | | iterations | 126 | | time_elapsed | 1097 | | total_timesteps | 2064384 | -------------------------------- ----------------------------------------- | time/ | | | fps | 1886 | | iterations | 127 | | time_elapsed | 1102 | | total_timesteps | 2080768 | | train/ | | | approx_kl | 0.006960577 | | clip_fraction | 0.0689 | | clip_range | 0.2 | | entropy_loss | -3.07 | | explained_variance | 0.917 | | learning_rate | 0.0003 | | loss | -0.0302 | | n_updates | 1260 | | policy_gradient_loss | -0.00248 | | std | 1.12 | | value_loss | 0.00841 | ----------------------------------------- ----------------------------------------- | time/ | | | fps | 1892 | | iterations | 128 | | time_elapsed | 1108 | | total_timesteps | 2097152 | | train/ | | | approx_kl | 0.007300884 | | clip_fraction | 0.0705 | | clip_range | 0.2 | | entropy_loss | -3.09 | | explained_variance | 0.915 | | learning_rate | 0.0003 | | loss | -0.0338 | | n_updates | 1270 | | policy_gradient_loss | -0.00351 | | std | 1.14 | | value_loss | 0.00336 | ----------------------------------------- Eval num_timesteps=2100000, episode_reward=37.33 +/- 41.91 Episode length: 2000.00 +/- 0.00 ----------------------------------------- | eval/ | | | mean_ep_length | 2e+03 | | mean_reward | 37.3 | | time/ | | | total_timesteps | 2100000 | | train/ | | | approx_kl | 0.007571588 | | clip_fraction | 0.076 | | clip_range | 0.2 | | entropy_loss | -3.1 | | explained_variance | 0.907 | | learning_rate | 0.0003 | | loss | -0.0278 | | n_updates | 1280 | | policy_gradient_loss | -0.00336 | | std | 1.14 | | value_loss | 0.00228 | ----------------------------------------- -------------------------------- | time/ | | | fps | 1878 | | iterations | 129 | | time_elapsed | 1124 | | total_timesteps | 2113536 | -------------------------------- ----------------------------------------- | time/ | | | fps | 1884 | | iterations | 130 | | time_elapsed | 1130 | | total_timesteps | 2129920 | | train/ | | | approx_kl | 0.007885255 | | clip_fraction | 0.088 | | clip_range | 0.2 | | entropy_loss | -3.11 | | explained_variance | 0.939 | | learning_rate | 0.0003 | | loss | -0.0388 | | n_updates | 1290 | | policy_gradient_loss | -0.00498 | | std | 1.15 | | value_loss | 0.00231 | ----------------------------------------- ------------------------------------------ | time/ | | | fps | 1890 | | iterations | 131 | | time_elapsed | 1135 | | total_timesteps | 2146304 | | train/ | | | approx_kl | 0.0073760273 | | clip_fraction | 0.0769 | | clip_range | 0.2 | | entropy_loss | -3.11 | | explained_variance | 0.955 | | learning_rate | 0.0003 | | loss | -0.0277 | | n_updates | 1300 | | policy_gradient_loss | -0.00306 | | std | 1.15 | | value_loss | 0.00294 | ------------------------------------------ Eval num_timesteps=2150000, episode_reward=31.84 +/- 38.92 Episode length: 2000.00 +/- 0.00 ----------------------------------------- | eval/ | | | mean_ep_length | 2e+03 | | mean_reward | 31.8 | | time/ | | | total_timesteps | 2150000 | | train/ | | | approx_kl | 0.006736047 | | clip_fraction | 0.0685 | | clip_range | 0.2 | | entropy_loss | -3.12 | | explained_variance | 0.913 | | learning_rate | 0.0003 | | loss | -0.0302 | | n_updates | 1310 | | policy_gradient_loss | -0.0021 | | std | 1.16 | | value_loss | 0.00422 | ----------------------------------------- -------------------------------- | time/ | | | fps | 1872 | | iterations | 132 | | time_elapsed | 1155 | | total_timesteps | 2162688 | -------------------------------- ----------------------------------------- | time/ | | | fps | 1878 | | iterations | 133 | | time_elapsed | 1160 | | total_timesteps | 2179072 | | train/ | | | approx_kl | 0.006166819 | | clip_fraction | 0.0668 | | clip_range | 0.2 | | entropy_loss | -3.13 | | explained_variance | 0.956 | | learning_rate | 0.0003 | | loss | -0.0473 | | n_updates | 1320 | | policy_gradient_loss | -0.00364 | | std | 1.16 | | value_loss | 0.00158 | ----------------------------------------- ------------------------------------------ | time/ | | | fps | 1884 | | iterations | 134 | | time_elapsed | 1165 | | total_timesteps | 2195456 | | train/ | | | approx_kl | 0.0075986157 | | clip_fraction | 0.0769 | | clip_range | 0.2 | | entropy_loss | -3.14 | | explained_variance | 0.966 | | learning_rate | 0.0003 | | loss | -0.0317 | | n_updates | 1330 | | policy_gradient_loss | -0.00398 | | std | 1.17 | | value_loss | 0.00307 | ------------------------------------------ Eval num_timesteps=2200000, episode_reward=26.98 +/- 37.84 Episode length: 2000.00 +/- 0.00 ----------------------------------------- | eval/ | | | mean_ep_length | 2e+03 | | mean_reward | 27 | | time/ | | | total_timesteps | 2200000 | | train/ | | | approx_kl | 0.008170303 | | clip_fraction | 0.0981 | | clip_range | 0.2 | | entropy_loss | -3.14 | | explained_variance | 0.964 | | learning_rate | 0.0003 | | loss | -0.0326 | | n_updates | 1340 | | policy_gradient_loss | -0.00415 | | std | 1.16 | | value_loss | 0.00349 | ----------------------------------------- [Diag @ 2,200,000 | n_sheep=3 | success=0%] NEVER_COMPACT 16/20 COMPACT_CANT_DRIVE 4/20 action_mag mean=0.067 p10=0.003 p90=0.067 (0=stopped, 1=full speed) min_flock_radius mean=7.25m best=1.61m (target <5m to compact) min_dog_to_com mean=0.97m best=0.20m (FLEE_DIST=7m) min_com_to_pen mean=13.28m best=5.53m reward/step (mean): progress=+0.0007 alignment=+0.0353 pen_bonus=+0.0008 step_cost=-0.0200 complete=+0.0000 -------------------------------- | time/ | | | fps | 1832 | | iterations | 135 | | time_elapsed | 1206 | | total_timesteps | 2211840 | -------------------------------- ----------------------------------------- | time/ | | | fps | 1805 | | iterations | 136 | | time_elapsed | 1234 | | total_timesteps | 2228224 | | train/ | | | approx_kl | 0.006131858 | | clip_fraction | 0.067 | | clip_range | 0.2 | | entropy_loss | -3.13 | | explained_variance | 0.927 | | learning_rate | 0.0003 | | loss | -0.0328 | | n_updates | 1350 | | policy_gradient_loss | -0.0022 | | std | 1.16 | | value_loss | 0.000981 | ----------------------------------------- ------------------------------------------ | time/ | | | fps | 1811 | | iterations | 137 | | time_elapsed | 1239 | | total_timesteps | 2244608 | | train/ | | | approx_kl | 0.0071705403 | | clip_fraction | 0.0699 | | clip_range | 0.2 | | entropy_loss | -3.12 | | explained_variance | 0.913 | | learning_rate | 0.0003 | | loss | -0.0391 | | n_updates | 1360 | | policy_gradient_loss | -0.0032 | | std | 1.15 | | value_loss | 0.00639 | ------------------------------------------ Eval num_timesteps=2250000, episode_reward=28.55 +/- 29.67 Episode length: 2000.00 +/- 0.00 ----------------------------------------- | eval/ | | | mean_ep_length | 2e+03 | | mean_reward | 28.5 | | time/ | | | total_timesteps | 2250000 | | train/ | | | approx_kl | 0.007929602 | | clip_fraction | 0.0812 | | clip_range | 0.2 | | entropy_loss | -3.14 | | explained_variance | 0.933 | | learning_rate | 0.0003 | | loss | -0.0592 | | n_updates | 1370 | | policy_gradient_loss | -0.00434 | | std | 1.17 | | value_loss | 0.00337 | ----------------------------------------- -------------------------------- | time/ | | | fps | 1800 | | iterations | 138 | | time_elapsed | 1255 | | total_timesteps | 2260992 | -------------------------------- ------------------------------------------ | time/ | | | fps | 1806 | | iterations | 139 | | time_elapsed | 1260 | | total_timesteps | 2277376 | | train/ | | | approx_kl | 0.0062256474 | | clip_fraction | 0.0592 | | clip_range | 0.2 | | entropy_loss | -3.15 | | explained_variance | 0.935 | | learning_rate | 0.0003 | | loss | -0.0368 | | n_updates | 1380 | | policy_gradient_loss | -0.00242 | | std | 1.17 | | value_loss | 0.00787 | ------------------------------------------ ------------------------------------------ | time/ | | | fps | 1812 | | iterations | 140 | | time_elapsed | 1265 | | total_timesteps | 2293760 | | train/ | | | approx_kl | 0.0075241774 | | clip_fraction | 0.0885 | | clip_range | 0.2 | | entropy_loss | -3.14 | | explained_variance | 0.948 | | learning_rate | 0.0003 | | loss | -0.0385 | | n_updates | 1390 | | policy_gradient_loss | -0.00346 | | std | 1.16 | | value_loss | 0.00172 | ------------------------------------------ Eval num_timesteps=2300000, episode_reward=43.34 +/- 34.73 Episode length: 2000.00 +/- 0.00 ------------------------------------------ | eval/ | | | mean_ep_length | 2e+03 | | mean_reward | 43.3 | | time/ | | | total_timesteps | 2300000 | | train/ | | | approx_kl | 0.0073855575 | | clip_fraction | 0.0753 | | clip_range | 0.2 | | entropy_loss | -3.12 | | explained_variance | 0.911 | | learning_rate | 0.0003 | | loss | -0.0377 | | n_updates | 1400 | | policy_gradient_loss | -0.0034 | | std | 1.15 | | value_loss | 0.00645 | ------------------------------------------ -------------------------------- | time/ | | | fps | 1801 | | iterations | 141 | | time_elapsed | 1282 | | total_timesteps | 2310144 | -------------------------------- ----------------------------------------- | time/ | | | fps | 1806 | | iterations | 142 | | time_elapsed | 1287 | | total_timesteps | 2326528 | | train/ | | | approx_kl | 0.007232903 | | clip_fraction | 0.0845 | | clip_range | 0.2 | | entropy_loss | -3.13 | | explained_variance | 0.956 | | learning_rate | 0.0003 | | loss | -0.0346 | | n_updates | 1410 | | policy_gradient_loss | -0.003 | | std | 1.16 | | value_loss | 0.00134 | ----------------------------------------- ----------------------------------------- | time/ | | | fps | 1812 | | iterations | 143 | | time_elapsed | 1292 | | total_timesteps | 2342912 | | train/ | | | approx_kl | 0.007283367 | | clip_fraction | 0.0785 | | clip_range | 0.2 | | entropy_loss | -3.14 | | explained_variance | 0.913 | | learning_rate | 0.0003 | | loss | -0.0306 | | n_updates | 1420 | | policy_gradient_loss | -0.00368 | | std | 1.17 | | value_loss | 0.00385 | ----------------------------------------- Eval num_timesteps=2350000, episode_reward=33.49 +/- 34.79 Episode length: 2000.00 +/- 0.00 ----------------------------------------- | eval/ | | | mean_ep_length | 2e+03 | | mean_reward | 33.5 | | time/ | | | total_timesteps | 2350000 | | train/ | | | approx_kl | 0.006632698 | | clip_fraction | 0.0647 | | clip_range | 0.2 | | entropy_loss | -3.15 | | explained_variance | 0.934 | | learning_rate | 0.0003 | | loss | -0.0469 | | n_updates | 1430 | | policy_gradient_loss | -0.00327 | | std | 1.17 | | value_loss | 0.00793 | ----------------------------------------- -------------------------------- | time/ | | | fps | 1800 | | iterations | 144 | | time_elapsed | 1310 | | total_timesteps | 2359296 | -------------------------------- ----------------------------------------- | time/ | | | fps | 1805 | | iterations | 145 | | time_elapsed | 1315 | | total_timesteps | 2375680 | | train/ | | | approx_kl | 0.008364577 | | clip_fraction | 0.089 | | clip_range | 0.2 | | entropy_loss | -3.15 | | explained_variance | 0.957 | | learning_rate | 0.0003 | | loss | -0.0464 | | n_updates | 1440 | | policy_gradient_loss | -0.00453 | | std | 1.17 | | value_loss | 0.00507 | ----------------------------------------- ----------------------------------------- | time/ | | | fps | 1810 | | iterations | 146 | | time_elapsed | 1321 | | total_timesteps | 2392064 | | train/ | | | approx_kl | 0.007854694 | | clip_fraction | 0.0927 | | clip_range | 0.2 | | entropy_loss | -3.15 | | explained_variance | 0.953 | | learning_rate | 0.0003 | | loss | -0.0436 | | n_updates | 1450 | | policy_gradient_loss | -0.00519 | | std | 1.17 | | value_loss | 0.00289 | ----------------------------------------- Eval num_timesteps=2400000, episode_reward=34.64 +/- 37.27 Episode length: 2000.00 +/- 0.00 ------------------------------------------ | eval/ | | | mean_ep_length | 2e+03 | | mean_reward | 34.6 | | time/ | | | total_timesteps | 2400000 | | train/ | | | approx_kl | 0.0076201856 | | clip_fraction | 0.0844 | | clip_range | 0.2 | | entropy_loss | -3.15 | | explained_variance | 0.945 | | learning_rate | 0.0003 | | loss | -0.0431 | | n_updates | 1460 | | policy_gradient_loss | -0.00554 | | std | 1.17 | | value_loss | 0.00196 | ------------------------------------------ [Diag @ 2,400,000 | n_sheep=3 | success=0%] NEVER_COMPACT 15/20 COMPACT_CANT_DRIVE 5/20 action_mag mean=0.058 p10=0.006 p90=0.053 (0=stopped, 1=full speed) min_flock_radius mean=6.68m best=0.96m (target <5m to compact) min_dog_to_com mean=0.92m best=0.16m (FLEE_DIST=7m) min_com_to_pen mean=12.18m best=5.62m reward/step (mean): progress=+0.0034 alignment=+0.0352 pen_bonus=+0.0010 step_cost=-0.0200 complete=+0.0000 [Curriculum] leaving stage n_sheep=3 after 800,000 steps | training success rate (last 100 eps) = 0% [Curriculum] → 4 sheep at step 2,400,000 -------------------------------- | time/ | | | fps | 1788 | | iterations | 147 | | time_elapsed | 1346 | | total_timesteps | 2408448 | -------------------------------- ----------------------------------------- | time/ | | | fps | 1794 | | iterations | 148 | | time_elapsed | 1351 | | total_timesteps | 2424832 | | train/ | | | approx_kl | 0.006801254 | | clip_fraction | 0.0797 | | clip_range | 0.2 | | entropy_loss | -3.16 | | explained_variance | 0.922 | | learning_rate | 0.0003 | | loss | -0.0313 | | n_updates | 1470 | | policy_gradient_loss | -0.00418 | | std | 1.18 | | value_loss | 0.00724 | ----------------------------------------- ----------------------------------------- | time/ | | | fps | 1798 | | iterations | 149 | | time_elapsed | 1357 | | total_timesteps | 2441216 | | train/ | | | approx_kl | 0.007604986 | | clip_fraction | 0.0758 | | clip_range | 0.2 | | entropy_loss | -3.18 | | explained_variance | 0.937 | | learning_rate | 0.0003 | | loss | -0.0354 | | n_updates | 1480 | | policy_gradient_loss | -0.00189 | | std | 1.19 | | value_loss | 0.00591 | ----------------------------------------- Eval num_timesteps=2450000, episode_reward=27.82 +/- 47.76 Episode length: 2000.00 +/- 0.00 ------------------------------------------ | eval/ | | | mean_ep_length | 2e+03 | | mean_reward | 27.8 | | time/ | | | total_timesteps | 2450000 | | train/ | | | approx_kl | 0.0070674624 | | clip_fraction | 0.0749 | | clip_range | 0.2 | | entropy_loss | -3.2 | | explained_variance | 0.893 | | learning_rate | 0.0003 | | loss | -0.0327 | | n_updates | 1490 | | policy_gradient_loss | -0.00322 | | std | 1.2 | | value_loss | 0.0105 | ------------------------------------------ -------------------------------- | time/ | | | fps | 1788 | | iterations | 150 | | time_elapsed | 1374 | | total_timesteps | 2457600 | -------------------------------- ----------------------------------------- | time/ | | | fps | 1792 | | iterations | 151 | | time_elapsed | 1380 | | total_timesteps | 2473984 | | train/ | | | approx_kl | 0.008372683 | | clip_fraction | 0.0874 | | clip_range | 0.2 | | entropy_loss | -3.21 | | explained_variance | 0.932 | | learning_rate | 0.0003 | | loss | -0.0381 | | n_updates | 1500 | | policy_gradient_loss | -0.00471 | | std | 1.21 | | value_loss | 0.00563 | ----------------------------------------- ----------------------------------------- | time/ | | | fps | 1796 | | iterations | 152 | | time_elapsed | 1385 | | total_timesteps | 2490368 | | train/ | | | approx_kl | 0.007761459 | | clip_fraction | 0.0794 | | clip_range | 0.2 | | entropy_loss | -3.23 | | explained_variance | 0.929 | | learning_rate | 0.0003 | | loss | -0.0345 | | n_updates | 1510 | | policy_gradient_loss | -0.00402 | | std | 1.22 | | value_loss | 0.00736 | ----------------------------------------- Eval num_timesteps=2500000, episode_reward=25.79 +/- 28.60 Episode length: 2000.00 +/- 0.00 ------------------------------------------ | eval/ | | | mean_ep_length | 2e+03 | | mean_reward | 25.8 | | time/ | | | total_timesteps | 2500000 | | train/ | | | approx_kl | 0.0070840344 | | clip_fraction | 0.0711 | | clip_range | 0.2 | | entropy_loss | -3.22 | | explained_variance | 0.9 | | learning_rate | 0.0003 | | loss | -0.0322 | | n_updates | 1520 | | policy_gradient_loss | -0.00397 | | std | 1.21 | | value_loss | 0.00517 | ------------------------------------------ -------------------------------- | time/ | | | fps | 1785 | | iterations | 153 | | time_elapsed | 1404 | | total_timesteps | 2506752 | -------------------------------- ------------------------------------------ | time/ | | | fps | 1788 | | iterations | 154 | | time_elapsed | 1410 | | total_timesteps | 2523136 | | train/ | | | approx_kl | 0.0062630484 | | clip_fraction | 0.069 | | clip_range | 0.2 | | entropy_loss | -3.22 | | explained_variance | 0.93 | | learning_rate | 0.0003 | | loss | -0.0363 | | n_updates | 1530 | | policy_gradient_loss | -0.00382 | | std | 1.21 | | value_loss | 0.00546 | ------------------------------------------ ----------------------------------------- | time/ | | | fps | 1792 | | iterations | 155 | | time_elapsed | 1416 | | total_timesteps | 2539520 | | train/ | | | approx_kl | 0.007609036 | | clip_fraction | 0.0815 | | clip_range | 0.2 | | entropy_loss | -3.23 | | explained_variance | 0.832 | | learning_rate | 0.0003 | | loss | -0.0404 | | n_updates | 1540 | | policy_gradient_loss | -0.00347 | | std | 1.22 | | value_loss | 0.00902 | ----------------------------------------- Eval num_timesteps=2550000, episode_reward=26.76 +/- 38.76 Episode length: 2000.00 +/- 0.00 ------------------------------------------ | eval/ | | | mean_ep_length | 2e+03 | | mean_reward | 26.8 | | time/ | | | total_timesteps | 2550000 | | train/ | | | approx_kl | 0.0070117847 | | clip_fraction | 0.0808 | | clip_range | 0.2 | | entropy_loss | -3.23 | | explained_variance | 0.863 | | learning_rate | 0.0003 | | loss | -0.0357 | | n_updates | 1550 | | policy_gradient_loss | -0.00279 | | std | 1.22 | | value_loss | 0.0114 | ------------------------------------------ -------------------------------- | time/ | | | fps | 1780 | | iterations | 156 | | time_elapsed | 1435 | | total_timesteps | 2555904 | -------------------------------- ------------------------------------------ | time/ | | | fps | 1786 | | iterations | 157 | | time_elapsed | 1440 | | total_timesteps | 2572288 | | train/ | | | approx_kl | 0.0070258966 | | clip_fraction | 0.0817 | | clip_range | 0.2 | | entropy_loss | -3.23 | | explained_variance | 0.941 | | learning_rate | 0.0003 | | loss | -0.039 | | n_updates | 1560 | | policy_gradient_loss | -0.00488 | | std | 1.22 | | value_loss | 0.00696 | ------------------------------------------ ----------------------------------------- | time/ | | | fps | 1791 | | iterations | 158 | | time_elapsed | 1445 | | total_timesteps | 2588672 | | train/ | | | approx_kl | 0.007600763 | | clip_fraction | 0.0842 | | clip_range | 0.2 | | entropy_loss | -3.23 | | explained_variance | 0.912 | | learning_rate | 0.0003 | | loss | -0.0363 | | n_updates | 1570 | | policy_gradient_loss | -0.00544 | | std | 1.22 | | value_loss | 0.00556 | ----------------------------------------- Eval num_timesteps=2600000, episode_reward=19.53 +/- 46.34 Episode length: 2000.00 +/- 0.00 ---------------------------------------- | eval/ | | | mean_ep_length | 2e+03 | | mean_reward | 19.5 | | time/ | | | total_timesteps | 2600000 | | train/ | | | approx_kl | 0.00714178 | | clip_fraction | 0.0783 | | clip_range | 0.2 | | entropy_loss | -3.23 | | explained_variance | 0.92 | | learning_rate | 0.0003 | | loss | -0.0352 | | n_updates | 1580 | | policy_gradient_loss | -0.00468 | | std | 1.22 | | value_loss | 0.00364 | ---------------------------------------- [Diag @ 2,600,000 | n_sheep=4 | success=0%] NEVER_COMPACT 20/20 action_mag mean=0.061 p10=0.006 p90=0.047 (0=stopped, 1=full speed) min_flock_radius mean=7.84m best=5.75m (target <5m to compact) min_dog_to_com mean=0.66m best=0.09m (FLEE_DIST=7m) min_com_to_pen mean=12.60m best=6.52m reward/step (mean): progress=-0.0028 alignment=+0.0337 pen_bonus=+0.0005 step_cost=-0.0200 complete=+0.0000 -------------------------------- | time/ | | | fps | 1768 | | iterations | 159 | | time_elapsed | 1473 | | total_timesteps | 2605056 | -------------------------------- ---------------------------------------- | time/ | | | fps | 1771 | | iterations | 160 | | time_elapsed | 1479 | | total_timesteps | 2621440 | | train/ | | | approx_kl | 0.00681924 | | clip_fraction | 0.0779 | | clip_range | 0.2 | | entropy_loss | -3.23 | | explained_variance | 0.946 | | learning_rate | 0.0003 | | loss | -0.0409 | | n_updates | 1590 | | policy_gradient_loss | -0.00346 | | std | 1.22 | | value_loss | 0.00377 | ---------------------------------------- ----------------------------------------- | time/ | | | fps | 1775 | | iterations | 161 | | time_elapsed | 1485 | | total_timesteps | 2637824 | | train/ | | | approx_kl | 0.008016385 | | clip_fraction | 0.0888 | | clip_range | 0.2 | | entropy_loss | -3.24 | | explained_variance | 0.931 | | learning_rate | 0.0003 | | loss | -0.0311 | | n_updates | 1600 | | policy_gradient_loss | -0.00526 | | std | 1.22 | | value_loss | 0.00681 | ----------------------------------------- Eval num_timesteps=2650000, episode_reward=28.98 +/- 40.07 Episode length: 2000.00 +/- 0.00 ----------------------------------------- | eval/ | | | mean_ep_length | 2e+03 | | mean_reward | 29 | | time/ | | | total_timesteps | 2650000 | | train/ | | | approx_kl | 0.006836592 | | clip_fraction | 0.0778 | | clip_range | 0.2 | | entropy_loss | -3.24 | | explained_variance | 0.9 | | learning_rate | 0.0003 | | loss | -0.0304 | | n_updates | 1610 | | policy_gradient_loss | -0.00255 | | std | 1.23 | | value_loss | 0.00574 | ----------------------------------------- -------------------------------- | time/ | | | fps | 1765 | | iterations | 162 | | time_elapsed | 1503 | | total_timesteps | 2654208 | -------------------------------- ------------------------------------------ | time/ | | | fps | 1770 | | iterations | 163 | | time_elapsed | 1508 | | total_timesteps | 2670592 | | train/ | | | approx_kl | 0.0072684484 | | clip_fraction | 0.0764 | | clip_range | 0.2 | | entropy_loss | -3.23 | | explained_variance | 0.948 | | learning_rate | 0.0003 | | loss | -0.0295 | | n_updates | 1620 | | policy_gradient_loss | -0.00325 | | std | 1.22 | | value_loss | 0.00254 | ------------------------------------------ ----------------------------------------- | time/ | | | fps | 1775 | | iterations | 164 | | time_elapsed | 1513 | | total_timesteps | 2686976 | | train/ | | | approx_kl | 0.007457966 | | clip_fraction | 0.0845 | | clip_range | 0.2 | | entropy_loss | -3.23 | | explained_variance | 0.919 | | learning_rate | 0.0003 | | loss | -0.0473 | | n_updates | 1630 | | policy_gradient_loss | -0.00505 | | std | 1.22 | | value_loss | 0.004 | ----------------------------------------- Eval num_timesteps=2700000, episode_reward=33.96 +/- 32.11 Episode length: 2000.00 +/- 0.00 ---------------------------------------- | eval/ | | | mean_ep_length | 2e+03 | | mean_reward | 34 | | time/ | | | total_timesteps | 2700000 | | train/ | | | approx_kl | 0.00796853 | | clip_fraction | 0.0782 | | clip_range | 0.2 | | entropy_loss | -3.22 | | explained_variance | 0.959 | | learning_rate | 0.0003 | | loss | -0.0336 | | n_updates | 1640 | | policy_gradient_loss | -0.00288 | | std | 1.21 | | value_loss | 0.00235 | ---------------------------------------- -------------------------------- | time/ | | | fps | 1761 | | iterations | 165 | | time_elapsed | 1534 | | total_timesteps | 2703360 | -------------------------------- ------------------------------------------ | time/ | | | fps | 1764 | | iterations | 166 | | time_elapsed | 1541 | | total_timesteps | 2719744 | | train/ | | | approx_kl | 0.0073700505 | | clip_fraction | 0.0857 | | clip_range | 0.2 | | entropy_loss | -3.21 | | explained_variance | 0.875 | | learning_rate | 0.0003 | | loss | -0.0255 | | n_updates | 1650 | | policy_gradient_loss | -0.00495 | | std | 1.21 | | value_loss | 0.00846 | ------------------------------------------ ----------------------------------------- | time/ | | | fps | 1768 | | iterations | 167 | | time_elapsed | 1546 | | total_timesteps | 2736128 | | train/ | | | approx_kl | 0.007965144 | | clip_fraction | 0.0858 | | clip_range | 0.2 | | entropy_loss | -3.22 | | explained_variance | 0.898 | | learning_rate | 0.0003 | | loss | -0.0451 | | n_updates | 1660 | | policy_gradient_loss | -0.00518 | | std | 1.22 | | value_loss | 0.00395 | ----------------------------------------- Eval num_timesteps=2750000, episode_reward=23.58 +/- 34.37 Episode length: 2000.00 +/- 0.00 ------------------------------------------ | eval/ | | | mean_ep_length | 2e+03 | | mean_reward | 23.6 | | time/ | | | total_timesteps | 2750000 | | train/ | | | approx_kl | 0.0065765316 | | clip_fraction | 0.0682 | | clip_range | 0.2 | | entropy_loss | -3.23 | | explained_variance | 0.934 | | learning_rate | 0.0003 | | loss | -0.0429 | | n_updates | 1670 | | policy_gradient_loss | -0.00379 | | std | 1.23 | | value_loss | 0.00677 | ------------------------------------------ -------------------------------- | time/ | | | fps | 1756 | | iterations | 168 | | time_elapsed | 1566 | | total_timesteps | 2752512 | -------------------------------- ------------------------------------------ | time/ | | | fps | 1761 | | iterations | 169 | | time_elapsed | 1571 | | total_timesteps | 2768896 | | train/ | | | approx_kl | 0.0066236854 | | clip_fraction | 0.0619 | | clip_range | 0.2 | | entropy_loss | -3.25 | | explained_variance | 0.935 | | learning_rate | 0.0003 | | loss | -0.0365 | | n_updates | 1680 | | policy_gradient_loss | -0.00239 | | std | 1.23 | | value_loss | 0.00922 | ------------------------------------------ ----------------------------------------- | time/ | | | fps | 1766 | | iterations | 170 | | time_elapsed | 1576 | | total_timesteps | 2785280 | | train/ | | | approx_kl | 0.007887056 | | clip_fraction | 0.0836 | | clip_range | 0.2 | | entropy_loss | -3.25 | | explained_variance | 0.899 | | learning_rate | 0.0003 | | loss | -0.0353 | | n_updates | 1690 | | policy_gradient_loss | -0.0053 | | std | 1.24 | | value_loss | 0.00635 | ----------------------------------------- Eval num_timesteps=2800000, episode_reward=33.57 +/- 35.56 Episode length: 2000.00 +/- 0.00 ------------------------------------------ | eval/ | | | mean_ep_length | 2e+03 | | mean_reward | 33.6 | | time/ | | | total_timesteps | 2800000 | | train/ | | | approx_kl | 0.0067548407 | | clip_fraction | 0.0804 | | clip_range | 0.2 | | entropy_loss | -3.25 | | explained_variance | 0.887 | | learning_rate | 0.0003 | | loss | -0.0408 | | n_updates | 1700 | | policy_gradient_loss | -0.00444 | | std | 1.24 | | value_loss | 0.0101 | ------------------------------------------ [Diag @ 2,800,000 | n_sheep=4 | success=0%] NEVER_COMPACT 19/20 COMPACT_CANT_DRIVE 1/20 action_mag mean=0.050 p10=0.003 p90=0.039 (0=stopped, 1=full speed) min_flock_radius mean=8.42m best=4.84m (target <5m to compact) min_dog_to_com mean=0.73m best=0.12m (FLEE_DIST=7m) min_com_to_pen mean=14.29m best=7.66m reward/step (mean): progress=-0.0027 alignment=+0.0365 pen_bonus=+0.0005 step_cost=-0.0200 complete=+0.0000 -------------------------------- | time/ | | | fps | 1746 | | iterations | 171 | | time_elapsed | 1604 | | total_timesteps | 2801664 | -------------------------------- ------------------------------------------ | time/ | | | fps | 1750 | | iterations | 172 | | time_elapsed | 1609 | | total_timesteps | 2818048 | | train/ | | | approx_kl | 0.0069283517 | | clip_fraction | 0.0847 | | clip_range | 0.2 | | entropy_loss | -3.24 | | explained_variance | 0.899 | | learning_rate | 0.0003 | | loss | -0.0476 | | n_updates | 1710 | | policy_gradient_loss | -0.00499 | | std | 1.23 | | value_loss | 0.00708 | ------------------------------------------ ----------------------------------------- | time/ | | | fps | 1754 | | iterations | 173 | | time_elapsed | 1615 | | total_timesteps | 2834432 | | train/ | | | approx_kl | 0.008303071 | | clip_fraction | 0.082 | | clip_range | 0.2 | | entropy_loss | -3.25 | | explained_variance | 0.911 | | learning_rate | 0.0003 | | loss | -0.0484 | | n_updates | 1720 | | policy_gradient_loss | -0.00388 | | std | 1.23 | | value_loss | 0.0061 | ----------------------------------------- Eval num_timesteps=2850000, episode_reward=34.42 +/- 32.01 Episode length: 2000.00 +/- 0.00 ------------------------------------------ | eval/ | | | mean_ep_length | 2e+03 | | mean_reward | 34.4 | | time/ | | | total_timesteps | 2850000 | | train/ | | | approx_kl | 0.0063731004 | | clip_fraction | 0.069 | | clip_range | 0.2 | | entropy_loss | -3.26 | | explained_variance | 0.951 | | learning_rate | 0.0003 | | loss | -0.029 | | n_updates | 1730 | | policy_gradient_loss | -0.00384 | | std | 1.25 | | value_loss | 0.00528 | ------------------------------------------ -------------------------------- | time/ | | | fps | 1745 | | iterations | 174 | | time_elapsed | 1633 | | total_timesteps | 2850816 | -------------------------------- ----------------------------------------- | time/ | | | fps | 1749 | | iterations | 175 | | time_elapsed | 1638 | | total_timesteps | 2867200 | | train/ | | | approx_kl | 0.008163793 | | clip_fraction | 0.0812 | | clip_range | 0.2 | | entropy_loss | -3.28 | | explained_variance | 0.935 | | learning_rate | 0.0003 | | loss | -0.0374 | | n_updates | 1740 | | policy_gradient_loss | -0.0032 | | std | 1.26 | | value_loss | 0.00432 | ----------------------------------------- ------------------------------------------ | time/ | | | fps | 1754 | | iterations | 176 | | time_elapsed | 1643 | | total_timesteps | 2883584 | | train/ | | | approx_kl | 0.0063439216 | | clip_fraction | 0.0743 | | clip_range | 0.2 | | entropy_loss | -3.29 | | explained_variance | 0.89 | | learning_rate | 0.0003 | | loss | -0.0372 | | n_updates | 1750 | | policy_gradient_loss | -0.00403 | | std | 1.26 | | value_loss | 0.00654 | ------------------------------------------ ----------------------------------------- | time/ | | | fps | 1759 | | iterations | 177 | | time_elapsed | 1648 | | total_timesteps | 2899968 | | train/ | | | approx_kl | 0.006967159 | | clip_fraction | 0.0761 | | clip_range | 0.2 | | entropy_loss | -3.29 | | explained_variance | 0.929 | | learning_rate | 0.0003 | | loss | -0.0462 | | n_updates | 1760 | | policy_gradient_loss | -0.00382 | | std | 1.26 | | value_loss | 0.00381 | ----------------------------------------- Eval num_timesteps=2900000, episode_reward=40.78 +/- 43.99 Episode length: 2000.00 +/- 0.00 ------------------------------------------ | eval/ | | | mean_ep_length | 2e+03 | | mean_reward | 40.8 | | time/ | | | total_timesteps | 2900000 | | train/ | | | approx_kl | 0.0075211767 | | clip_fraction | 0.0727 | | clip_range | 0.2 | | entropy_loss | -3.29 | | explained_variance | 0.955 | | learning_rate | 0.0003 | | loss | -0.0178 | | n_updates | 1770 | | policy_gradient_loss | -0.00285 | | std | 1.27 | | value_loss | 0.00798 | ------------------------------------------ -------------------------------- | time/ | | | fps | 1751 | | iterations | 178 | | time_elapsed | 1664 | | total_timesteps | 2916352 | -------------------------------- ----------------------------------------- | time/ | | | fps | 1756 | | iterations | 179 | | time_elapsed | 1669 | | total_timesteps | 2932736 | | train/ | | | approx_kl | 0.006763531 | | clip_fraction | 0.0678 | | clip_range | 0.2 | | entropy_loss | -3.3 | | explained_variance | 0.91 | | learning_rate | 0.0003 | | loss | -0.0349 | | n_updates | 1780 | | policy_gradient_loss | -0.00361 | | std | 1.27 | | value_loss | 0.00528 | ----------------------------------------- ------------------------------------------ | time/ | | | fps | 1760 | | iterations | 180 | | time_elapsed | 1675 | | total_timesteps | 2949120 | | train/ | | | approx_kl | 0.0067441636 | | clip_fraction | 0.0732 | | clip_range | 0.2 | | entropy_loss | -3.3 | | explained_variance | 0.888 | | learning_rate | 0.0003 | | loss | -0.0261 | | n_updates | 1790 | | policy_gradient_loss | -0.00291 | | std | 1.27 | | value_loss | 0.00582 | ------------------------------------------ Eval num_timesteps=2950000, episode_reward=48.39 +/- 31.91 Episode length: 2000.00 +/- 0.00 ------------------------------------------ | eval/ | | | mean_ep_length | 2e+03 | | mean_reward | 48.4 | | time/ | | | total_timesteps | 2950000 | | train/ | | | approx_kl | 0.0076025603 | | clip_fraction | 0.0858 | | clip_range | 0.2 | | entropy_loss | -3.31 | | explained_variance | 0.92 | | learning_rate | 0.0003 | | loss | -0.0394 | | n_updates | 1800 | | policy_gradient_loss | -0.00443 | | std | 1.27 | | value_loss | 0.00647 | ------------------------------------------ -------------------------------- | time/ | | | fps | 1751 | | iterations | 181 | | time_elapsed | 1693 | | total_timesteps | 2965504 | -------------------------------- ----------------------------------------- | time/ | | | fps | 1754 | | iterations | 182 | | time_elapsed | 1699 | | total_timesteps | 2981888 | | train/ | | | approx_kl | 0.008041672 | | clip_fraction | 0.0795 | | clip_range | 0.2 | | entropy_loss | -3.3 | | explained_variance | 0.939 | | learning_rate | 0.0003 | | loss | -0.0344 | | n_updates | 1810 | | policy_gradient_loss | -0.00456 | | std | 1.27 | | value_loss | 0.00404 | ----------------------------------------- ------------------------------------------ | time/ | | | fps | 1758 | | iterations | 183 | | time_elapsed | 1704 | | total_timesteps | 2998272 | | train/ | | | approx_kl | 0.0066829836 | | clip_fraction | 0.0712 | | clip_range | 0.2 | | entropy_loss | -3.32 | | explained_variance | 0.921 | | learning_rate | 0.0003 | | loss | -0.0361 | | n_updates | 1820 | | policy_gradient_loss | -0.00379 | | std | 1.28 | | value_loss | 0.00818 | ------------------------------------------ Eval num_timesteps=3000000, episode_reward=33.06 +/- 47.57 Episode length: 2000.00 +/- 0.00 ----------------------------------------- | eval/ | | | mean_ep_length | 2e+03 | | mean_reward | 33.1 | | time/ | | | total_timesteps | 3000000 | | train/ | | | approx_kl | 0.006152373 | | clip_fraction | 0.0633 | | clip_range | 0.2 | | entropy_loss | -3.33 | | explained_variance | 0.912 | | learning_rate | 0.0003 | | loss | -0.0316 | | n_updates | 1830 | | policy_gradient_loss | -0.00335 | | std | 1.29 | | value_loss | 0.00404 | ----------------------------------------- [Diag @ 3,000,000 | n_sheep=4 | success=0%] NEVER_COMPACT 20/20 action_mag mean=0.049 p10=0.005 p90=0.046 (0=stopped, 1=full speed) min_flock_radius mean=8.21m best=5.29m (target <5m to compact) min_dog_to_com mean=0.76m best=0.22m (FLEE_DIST=7m) min_com_to_pen mean=12.62m best=4.77m reward/step (mean): progress=+0.0089 alignment=+0.0386 pen_bonus=+0.0008 step_cost=-0.0200 complete=+0.0000 -------------------------------- | time/ | | | fps | 1740 | | iterations | 184 | | time_elapsed | 1731 | | total_timesteps | 3014656 | -------------------------------- ----------------------------------------- | time/ | | | fps | 1745 | | iterations | 185 | | time_elapsed | 1736 | | total_timesteps | 3031040 | | train/ | | | approx_kl | 0.006385569 | | clip_fraction | 0.0703 | | clip_range | 0.2 | | entropy_loss | -3.34 | | explained_variance | 0.919 | | learning_rate | 0.0003 | | loss | -0.0313 | | n_updates | 1840 | | policy_gradient_loss | -0.00274 | | std | 1.3 | | value_loss | 0.00503 | ----------------------------------------- ----------------------------------------- | time/ | | | fps | 1748 | | iterations | 186 | | time_elapsed | 1743 | | total_timesteps | 3047424 | | train/ | | | approx_kl | 0.007695101 | | clip_fraction | 0.0784 | | clip_range | 0.2 | | entropy_loss | -3.36 | | explained_variance | 0.935 | | learning_rate | 0.0003 | | loss | -0.0244 | | n_updates | 1850 | | policy_gradient_loss | -0.00342 | | std | 1.31 | | value_loss | 0.0051 | ----------------------------------------- Eval num_timesteps=3050000, episode_reward=45.25 +/- 31.57 Episode length: 2000.00 +/- 0.00 ------------------------------------------ | eval/ | | | mean_ep_length | 2e+03 | | mean_reward | 45.2 | | time/ | | | total_timesteps | 3050000 | | train/ | | | approx_kl | 0.0067556566 | | clip_fraction | 0.082 | | clip_range | 0.2 | | entropy_loss | -3.37 | | explained_variance | 0.868 | | learning_rate | 0.0003 | | loss | -0.0349 | | n_updates | 1860 | | policy_gradient_loss | -0.00353 | | std | 1.31 | | value_loss | 0.00931 | ------------------------------------------ -------------------------------- | time/ | | | fps | 1738 | | iterations | 187 | | time_elapsed | 1762 | | total_timesteps | 3063808 | -------------------------------- ----------------------------------------- | time/ | | | fps | 1741 | | iterations | 188 | | time_elapsed | 1768 | | total_timesteps | 3080192 | | train/ | | | approx_kl | 0.008263266 | | clip_fraction | 0.0792 | | clip_range | 0.2 | | entropy_loss | -3.36 | | explained_variance | 0.924 | | learning_rate | 0.0003 | | loss | -0.0411 | | n_updates | 1870 | | policy_gradient_loss | -0.00382 | | std | 1.31 | | value_loss | 0.00429 | ----------------------------------------- ----------------------------------------- | time/ | | | fps | 1746 | | iterations | 189 | | time_elapsed | 1773 | | total_timesteps | 3096576 | | train/ | | | approx_kl | 0.008488305 | | clip_fraction | 0.08 | | clip_range | 0.2 | | entropy_loss | -3.37 | | explained_variance | 0.925 | | learning_rate | 0.0003 | | loss | -0.0292 | | n_updates | 1880 | | policy_gradient_loss | -0.00441 | | std | 1.31 | | value_loss | 0.00748 | ----------------------------------------- Eval num_timesteps=3100000, episode_reward=30.63 +/- 33.70 Episode length: 2000.00 +/- 0.00 ------------------------------------------ | eval/ | | | mean_ep_length | 2e+03 | | mean_reward | 30.6 | | time/ | | | total_timesteps | 3100000 | | train/ | | | approx_kl | 0.0065515246 | | clip_fraction | 0.0736 | | clip_range | 0.2 | | entropy_loss | -3.35 | | explained_variance | 0.932 | | learning_rate | 0.0003 | | loss | 0.00192 | | n_updates | 1890 | | policy_gradient_loss | -0.00334 | | std | 1.3 | | value_loss | 0.00902 | ------------------------------------------ -------------------------------- | time/ | | | fps | 1737 | | iterations | 190 | | time_elapsed | 1791 | | total_timesteps | 3112960 | -------------------------------- ------------------------------------------ | time/ | | | fps | 1741 | | iterations | 191 | | time_elapsed | 1796 | | total_timesteps | 3129344 | | train/ | | | approx_kl | 0.0068135276 | | clip_fraction | 0.0721 | | clip_range | 0.2 | | entropy_loss | -3.35 | | explained_variance | 0.933 | | learning_rate | 0.0003 | | loss | -0.036 | | n_updates | 1900 | | policy_gradient_loss | -0.00403 | | std | 1.29 | | value_loss | 0.00616 | ------------------------------------------ ------------------------------------------ | time/ | | | fps | 1731 | | iterations | 192 | | time_elapsed | 1817 | | total_timesteps | 3145728 | | train/ | | | approx_kl | 0.0061126407 | | clip_fraction | 0.0615 | | clip_range | 0.2 | | entropy_loss | -3.35 | | explained_variance | 0.921 | | learning_rate | 0.0003 | | loss | -0.0355 | | n_updates | 1910 | | policy_gradient_loss | -0.00318 | | std | 1.3 | | value_loss | 0.0104 | ------------------------------------------ Eval num_timesteps=3150000, episode_reward=33.88 +/- 34.31 Episode length: 2000.00 +/- 0.00 ----------------------------------------- | eval/ | | | mean_ep_length | 2e+03 | | mean_reward | 33.9 | | time/ | | | total_timesteps | 3150000 | | train/ | | | approx_kl | 0.007734685 | | clip_fraction | 0.0778 | | clip_range | 0.2 | | entropy_loss | -3.35 | | explained_variance | 0.899 | | learning_rate | 0.0003 | | loss | -0.0323 | | n_updates | 1920 | | policy_gradient_loss | -0.00432 | | std | 1.3 | | value_loss | 0.0091 | ----------------------------------------- -------------------------------- | time/ | | | fps | 1714 | | iterations | 193 | | time_elapsed | 1844 | | total_timesteps | 3162112 | -------------------------------- ----------------------------------------- | time/ | | | fps | 1717 | | iterations | 194 | | time_elapsed | 1850 | | total_timesteps | 3178496 | | train/ | | | approx_kl | 0.007997783 | | clip_fraction | 0.0782 | | clip_range | 0.2 | | entropy_loss | -3.35 | | explained_variance | 0.91 | | learning_rate | 0.0003 | | loss | -0.0525 | | n_updates | 1930 | | policy_gradient_loss | -0.00523 | | std | 1.3 | | value_loss | 0.00283 | ----------------------------------------- ----------------------------------------- | time/ | | | fps | 1720 | | iterations | 195 | | time_elapsed | 1857 | | total_timesteps | 3194880 | | train/ | | | approx_kl | 0.007701534 | | clip_fraction | 0.0712 | | clip_range | 0.2 | | entropy_loss | -3.34 | | explained_variance | 0.927 | | learning_rate | 0.0003 | | loss | -0.0367 | | n_updates | 1940 | | policy_gradient_loss | -0.00288 | | std | 1.3 | | value_loss | 0.0126 | ----------------------------------------- Eval num_timesteps=3200000, episode_reward=46.55 +/- 34.01 Episode length: 2000.00 +/- 0.00 ----------------------------------------- | eval/ | | | mean_ep_length | 2e+03 | | mean_reward | 46.6 | | time/ | | | total_timesteps | 3200000 | | train/ | | | approx_kl | 0.006747664 | | clip_fraction | 0.0766 | | clip_range | 0.2 | | entropy_loss | -3.35 | | explained_variance | 0.93 | | learning_rate | 0.0003 | | loss | -0.0411 | | n_updates | 1950 | | policy_gradient_loss | -0.00404 | | std | 1.3 | | value_loss | 0.00409 | ----------------------------------------- [Diag @ 3,200,000 | n_sheep=4 | success=0%] NEVER_COMPACT 20/20 action_mag mean=0.078 p10=0.005 p90=0.057 (0=stopped, 1=full speed) min_flock_radius mean=8.76m best=6.32m (target <5m to compact) min_dog_to_com mean=0.81m best=0.36m (FLEE_DIST=7m) min_com_to_pen mean=13.75m best=6.91m reward/step (mean): progress=-0.0020 alignment=+0.0384 pen_bonus=+0.0003 step_cost=-0.0200 complete=+0.0000 [Curriculum] leaving stage n_sheep=4 after 800,000 steps | training success rate (last 100 eps) = 0% [Curriculum] → 5 sheep at step 3,200,000 -------------------------------- | time/ | | | fps | 1704 | | iterations | 196 | | time_elapsed | 1884 | | total_timesteps | 3211264 | -------------------------------- ------------------------------------------ | time/ | | | fps | 1707 | | iterations | 197 | | time_elapsed | 1889 | | total_timesteps | 3227648 | | train/ | | | approx_kl | 0.0068222135 | | clip_fraction | 0.0816 | | clip_range | 0.2 | | entropy_loss | -3.36 | | explained_variance | 0.922 | | learning_rate | 0.0003 | | loss | -0.0386 | | n_updates | 1960 | | policy_gradient_loss | -0.00374 | | std | 1.31 | | value_loss | 0.0112 | ------------------------------------------ ----------------------------------------- | time/ | | | fps | 1711 | | iterations | 198 | | time_elapsed | 1895 | | total_timesteps | 3244032 | | train/ | | | approx_kl | 0.006939999 | | clip_fraction | 0.0829 | | clip_range | 0.2 | | entropy_loss | -3.36 | | explained_variance | 0.955 | | learning_rate | 0.0003 | | loss | -0.0439 | | n_updates | 1970 | | policy_gradient_loss | -0.00433 | | std | 1.31 | | value_loss | 0.00895 | ----------------------------------------- Eval num_timesteps=3250000, episode_reward=21.19 +/- 37.18 Episode length: 2000.00 +/- 0.00 ----------------------------------------- | eval/ | | | mean_ep_length | 2e+03 | | mean_reward | 21.2 | | time/ | | | total_timesteps | 3250000 | | train/ | | | approx_kl | 0.007944042 | | clip_fraction | 0.0812 | | clip_range | 0.2 | | entropy_loss | -3.37 | | explained_variance | 0.925 | | learning_rate | 0.0003 | | loss | -0.0379 | | n_updates | 1980 | | policy_gradient_loss | -0.00306 | | std | 1.31 | | value_loss | 0.00578 | ----------------------------------------- -------------------------------- | time/ | | | fps | 1702 | | iterations | 199 | | time_elapsed | 1914 | | total_timesteps | 3260416 | -------------------------------- ----------------------------------------- | time/ | | | fps | 1706 | | iterations | 200 | | time_elapsed | 1920 | | total_timesteps | 3276800 | | train/ | | | approx_kl | 0.007009124 | | clip_fraction | 0.0786 | | clip_range | 0.2 | | entropy_loss | -3.36 | | explained_variance | 0.945 | | learning_rate | 0.0003 | | loss | -0.0398 | | n_updates | 1990 | | policy_gradient_loss | -0.00469 | | std | 1.31 | | value_loss | 0.00344 | ----------------------------------------- ----------------------------------------- | time/ | | | fps | 1709 | | iterations | 201 | | time_elapsed | 1926 | | total_timesteps | 3293184 | | train/ | | | approx_kl | 0.007446406 | | clip_fraction | 0.0736 | | clip_range | 0.2 | | entropy_loss | -3.36 | | explained_variance | 0.957 | | learning_rate | 0.0003 | | loss | -0.0493 | | n_updates | 2000 | | policy_gradient_loss | -0.00431 | | std | 1.31 | | value_loss | 0.00262 | ----------------------------------------- Eval num_timesteps=3300000, episode_reward=18.42 +/- 36.17 Episode length: 2000.00 +/- 0.00 ----------------------------------------- | eval/ | | | mean_ep_length | 2e+03 | | mean_reward | 18.4 | | time/ | | | total_timesteps | 3300000 | | train/ | | | approx_kl | 0.007855328 | | clip_fraction | 0.0783 | | clip_range | 0.2 | | entropy_loss | -3.38 | | explained_variance | 0.951 | | learning_rate | 0.0003 | | loss | -0.0381 | | n_updates | 2010 | | policy_gradient_loss | -0.00422 | | std | 1.32 | | value_loss | 0.00379 | ----------------------------------------- -------------------------------- | time/ | | | fps | 1701 | | iterations | 202 | | time_elapsed | 1945 | | total_timesteps | 3309568 | -------------------------------- ------------------------------------------ | time/ | | | fps | 1704 | | iterations | 203 | | time_elapsed | 1951 | | total_timesteps | 3325952 | | train/ | | | approx_kl | 0.0073990654 | | clip_fraction | 0.0773 | | clip_range | 0.2 | | entropy_loss | -3.38 | | explained_variance | 0.89 | | learning_rate | 0.0003 | | loss | -0.0319 | | n_updates | 2020 | | policy_gradient_loss | -0.00507 | | std | 1.32 | | value_loss | 0.0165 | ------------------------------------------ ------------------------------------------ | time/ | | | fps | 1707 | | iterations | 204 | | time_elapsed | 1956 | | total_timesteps | 3342336 | | train/ | | | approx_kl | 0.0076738494 | | clip_fraction | 0.0913 | | clip_range | 0.2 | | entropy_loss | -3.38 | | explained_variance | 0.914 | | learning_rate | 0.0003 | | loss | -0.0326 | | n_updates | 2030 | | policy_gradient_loss | -0.00611 | | std | 1.32 | | value_loss | 0.00854 | ------------------------------------------ Eval num_timesteps=3350000, episode_reward=39.75 +/- 38.09 Episode length: 2000.00 +/- 0.00 ----------------------------------------- | eval/ | | | mean_ep_length | 2e+03 | | mean_reward | 39.8 | | time/ | | | total_timesteps | 3350000 | | train/ | | | approx_kl | 0.007704767 | | clip_fraction | 0.0813 | | clip_range | 0.2 | | entropy_loss | -3.39 | | explained_variance | 0.822 | | learning_rate | 0.0003 | | loss | -0.0351 | | n_updates | 2040 | | policy_gradient_loss | -0.0056 | | std | 1.33 | | value_loss | 0.0095 | ----------------------------------------- -------------------------------- | time/ | | | fps | 1700 | | iterations | 205 | | time_elapsed | 1974 | | total_timesteps | 3358720 | -------------------------------- ----------------------------------------- | time/ | | | fps | 1703 | | iterations | 206 | | time_elapsed | 1980 | | total_timesteps | 3375104 | | train/ | | | approx_kl | 0.006841295 | | clip_fraction | 0.0682 | | clip_range | 0.2 | | entropy_loss | -3.39 | | explained_variance | 0.973 | | learning_rate | 0.0003 | | loss | -0.04 | | n_updates | 2050 | | policy_gradient_loss | -0.00457 | | std | 1.33 | | value_loss | 0.00456 | ----------------------------------------- ------------------------------------------ | time/ | | | fps | 1707 | | iterations | 207 | | time_elapsed | 1986 | | total_timesteps | 3391488 | | train/ | | | approx_kl | 0.0063885115 | | clip_fraction | 0.0749 | | clip_range | 0.2 | | entropy_loss | -3.4 | | explained_variance | 0.962 | | learning_rate | 0.0003 | | loss | -0.041 | | n_updates | 2060 | | policy_gradient_loss | -0.00455 | | std | 1.34 | | value_loss | 0.00373 | ------------------------------------------ Eval num_timesteps=3400000, episode_reward=26.62 +/- 43.12 Episode length: 2000.00 +/- 0.00 ----------------------------------------- | eval/ | | | mean_ep_length | 2e+03 | | mean_reward | 26.6 | | time/ | | | total_timesteps | 3400000 | | train/ | | | approx_kl | 0.006273965 | | clip_fraction | 0.0709 | | clip_range | 0.2 | | entropy_loss | -3.4 | | explained_variance | 0.956 | | learning_rate | 0.0003 | | loss | -0.0465 | | n_updates | 2070 | | policy_gradient_loss | -0.00249 | | std | 1.33 | | value_loss | 0.00679 | ----------------------------------------- [Diag @ 3,400,000 | n_sheep=5 | success=0%] NEVER_COMPACT 20/20 action_mag mean=0.089 p10=0.005 p90=0.074 (0=stopped, 1=full speed) min_flock_radius mean=9.14m best=5.59m (target <5m to compact) min_dog_to_com mean=0.69m best=0.10m (FLEE_DIST=7m) min_com_to_pen mean=12.77m best=5.15m reward/step (mean): progress=-0.0015 alignment=+0.0368 pen_bonus=+0.0020 step_cost=-0.0200 complete=+0.0000 -------------------------------- | time/ | | | fps | 1691 | | iterations | 208 | | time_elapsed | 2014 | | total_timesteps | 3407872 | -------------------------------- ----------------------------------------- | time/ | | | fps | 1695 | | iterations | 209 | | time_elapsed | 2019 | | total_timesteps | 3424256 | | train/ | | | approx_kl | 0.006433293 | | clip_fraction | 0.0727 | | clip_range | 0.2 | | entropy_loss | -3.4 | | explained_variance | 0.932 | | learning_rate | 0.0003 | | loss | -0.0268 | | n_updates | 2080 | | policy_gradient_loss | -0.00365 | | std | 1.33 | | value_loss | 0.00657 | ----------------------------------------- ----------------------------------------- | time/ | | | fps | 1698 | | iterations | 210 | | time_elapsed | 2025 | | total_timesteps | 3440640 | | train/ | | | approx_kl | 0.007235542 | | clip_fraction | 0.0839 | | clip_range | 0.2 | | entropy_loss | -3.39 | | explained_variance | 0.935 | | learning_rate | 0.0003 | | loss | -0.0344 | | n_updates | 2090 | | policy_gradient_loss | -0.00417 | | std | 1.32 | | value_loss | 0.0137 | ----------------------------------------- Eval num_timesteps=3450000, episode_reward=35.54 +/- 43.01 Episode length: 2000.00 +/- 0.00 ----------------------------------------- | eval/ | | | mean_ep_length | 2e+03 | | mean_reward | 35.5 | | time/ | | | total_timesteps | 3450000 | | train/ | | | approx_kl | 0.007782845 | | clip_fraction | 0.0859 | | clip_range | 0.2 | | entropy_loss | -3.4 | | explained_variance | 0.924 | | learning_rate | 0.0003 | | loss | -0.044 | | n_updates | 2100 | | policy_gradient_loss | -0.00561 | | std | 1.34 | | value_loss | 0.0043 | ----------------------------------------- -------------------------------- | time/ | | | fps | 1690 | | iterations | 211 | | time_elapsed | 2044 | | total_timesteps | 3457024 | -------------------------------- ------------------------------------------ | time/ | | | fps | 1693 | | iterations | 212 | | time_elapsed | 2050 | | total_timesteps | 3473408 | | train/ | | | approx_kl | 0.0075765867 | | clip_fraction | 0.0746 | | clip_range | 0.2 | | entropy_loss | -3.41 | | explained_variance | 0.896 | | learning_rate | 0.0003 | | loss | -0.0293 | | n_updates | 2110 | | policy_gradient_loss | -0.00406 | | std | 1.33 | | value_loss | 0.011 | ------------------------------------------ ------------------------------------------ | time/ | | | fps | 1696 | | iterations | 213 | | time_elapsed | 2056 | | total_timesteps | 3489792 | | train/ | | | approx_kl | 0.0072322125 | | clip_fraction | 0.071 | | clip_range | 0.2 | | entropy_loss | -3.41 | | explained_variance | 0.949 | | learning_rate | 0.0003 | | loss | -0.0498 | | n_updates | 2120 | | policy_gradient_loss | -0.00421 | | std | 1.34 | | value_loss | 0.006 | ------------------------------------------ Eval num_timesteps=3500000, episode_reward=54.69 +/- 47.39 Episode length: 2000.00 +/- 0.00 ------------------------------------------ | eval/ | | | mean_ep_length | 2e+03 | | mean_reward | 54.7 | | time/ | | | total_timesteps | 3500000 | | train/ | | | approx_kl | 0.0073479656 | | clip_fraction | 0.0778 | | clip_range | 0.2 | | entropy_loss | -3.4 | | explained_variance | 0.824 | | learning_rate | 0.0003 | | loss | -0.0408 | | n_updates | 2130 | | policy_gradient_loss | -0.00465 | | std | 1.32 | | value_loss | 0.00657 | ------------------------------------------ -------------------------------- | time/ | | | fps | 1688 | | iterations | 214 | | time_elapsed | 2076 | | total_timesteps | 3506176 | -------------------------------- ----------------------------------------- | time/ | | | fps | 1692 | | iterations | 215 | | time_elapsed | 2081 | | total_timesteps | 3522560 | | train/ | | | approx_kl | 0.007274649 | | clip_fraction | 0.0798 | | clip_range | 0.2 | | entropy_loss | -3.39 | | explained_variance | 0.951 | | learning_rate | 0.0003 | | loss | -0.0356 | | n_updates | 2140 | | policy_gradient_loss | -0.00383 | | std | 1.33 | | value_loss | 0.00355 | ----------------------------------------- ------------------------------------------ | time/ | | | fps | 1695 | | iterations | 216 | | time_elapsed | 2087 | | total_timesteps | 3538944 | | train/ | | | approx_kl | 0.0068056686 | | clip_fraction | 0.0726 | | clip_range | 0.2 | | entropy_loss | -3.38 | | explained_variance | 0.955 | | learning_rate | 0.0003 | | loss | -0.0428 | | n_updates | 2150 | | policy_gradient_loss | -0.00356 | | std | 1.32 | | value_loss | 0.00378 | ------------------------------------------ Eval num_timesteps=3550000, episode_reward=8.69 +/- 39.03 Episode length: 2000.00 +/- 0.00 ----------------------------------------- | eval/ | | | mean_ep_length | 2e+03 | | mean_reward | 8.69 | | time/ | | | total_timesteps | 3550000 | | train/ | | | approx_kl | 0.008211401 | | clip_fraction | 0.0801 | | clip_range | 0.2 | | entropy_loss | -3.38 | | explained_variance | 0.972 | | learning_rate | 0.0003 | | loss | -0.0366 | | n_updates | 2160 | | policy_gradient_loss | -0.00453 | | std | 1.32 | | value_loss | 0.00445 | ----------------------------------------- -------------------------------- | time/ | | | fps | 1687 | | iterations | 217 | | time_elapsed | 2106 | | total_timesteps | 3555328 | -------------------------------- ----------------------------------------- | time/ | | | fps | 1690 | | iterations | 218 | | time_elapsed | 2112 | | total_timesteps | 3571712 | | train/ | | | approx_kl | 0.008278061 | | clip_fraction | 0.0871 | | clip_range | 0.2 | | entropy_loss | -3.38 | | explained_variance | 0.931 | | learning_rate | 0.0003 | | loss | -0.0324 | | n_updates | 2170 | | policy_gradient_loss | -0.00486 | | std | 1.32 | | value_loss | 0.00377 | ----------------------------------------- ----------------------------------------- | time/ | | | fps | 1693 | | iterations | 219 | | time_elapsed | 2119 | | total_timesteps | 3588096 | | train/ | | | approx_kl | 0.007908824 | | clip_fraction | 0.0777 | | clip_range | 0.2 | | entropy_loss | -3.39 | | explained_variance | 0.951 | | learning_rate | 0.0003 | | loss | -0.0353 | | n_updates | 2180 | | policy_gradient_loss | -0.00318 | | std | 1.32 | | value_loss | 0.00768 | ----------------------------------------- Eval num_timesteps=3600000, episode_reward=26.00 +/- 35.20 Episode length: 2000.00 +/- 0.00 ------------------------------------------ | eval/ | | | mean_ep_length | 2e+03 | | mean_reward | 26 | | time/ | | | total_timesteps | 3600000 | | train/ | | | approx_kl | 0.0068260087 | | clip_fraction | 0.0761 | | clip_range | 0.2 | | entropy_loss | -3.38 | | explained_variance | 0.946 | | learning_rate | 0.0003 | | loss | -0.0257 | | n_updates | 2190 | | policy_gradient_loss | -0.00375 | | std | 1.32 | | value_loss | 0.00745 | ------------------------------------------ [Diag @ 3,600,000 | n_sheep=5 | success=0%] NEVER_COMPACT 20/20 action_mag mean=0.114 p10=0.006 p90=0.281 (0=stopped, 1=full speed) min_flock_radius mean=9.62m best=5.04m (target <5m to compact) min_dog_to_com mean=0.77m best=0.40m (FLEE_DIST=7m) min_com_to_pen mean=13.31m best=6.37m reward/step (mean): progress=+0.0071 alignment=+0.0385 pen_bonus=+0.0008 step_cost=-0.0200 complete=+0.0000 -------------------------------- | time/ | | | fps | 1677 | | iterations | 220 | | time_elapsed | 2148 | | total_timesteps | 3604480 | -------------------------------- ------------------------------------------ | time/ | | | fps | 1680 | | iterations | 221 | | time_elapsed | 2154 | | total_timesteps | 3620864 | | train/ | | | approx_kl | 0.0084966235 | | clip_fraction | 0.0849 | | clip_range | 0.2 | | entropy_loss | -3.38 | | explained_variance | 0.936 | | learning_rate | 0.0003 | | loss | -0.0498 | | n_updates | 2200 | | policy_gradient_loss | -0.00478 | | std | 1.32 | | value_loss | 0.00856 | ------------------------------------------ ----------------------------------------- | time/ | | | fps | 1683 | | iterations | 222 | | time_elapsed | 2160 | | total_timesteps | 3637248 | | train/ | | | approx_kl | 0.007236682 | | clip_fraction | 0.072 | | clip_range | 0.2 | | entropy_loss | -3.37 | | explained_variance | 0.956 | | learning_rate | 0.0003 | | loss | -0.0436 | | n_updates | 2210 | | policy_gradient_loss | -0.0054 | | std | 1.31 | | value_loss | 0.00748 | ----------------------------------------- Eval num_timesteps=3650000, episode_reward=48.26 +/- 45.24 Episode length: 2000.00 +/- 0.00 ------------------------------------------ | eval/ | | | mean_ep_length | 2e+03 | | mean_reward | 48.3 | | time/ | | | total_timesteps | 3650000 | | train/ | | | approx_kl | 0.0076099336 | | clip_fraction | 0.0694 | | clip_range | 0.2 | | entropy_loss | -3.37 | | explained_variance | 0.942 | | learning_rate | 0.0003 | | loss | -0.037 | | n_updates | 2220 | | policy_gradient_loss | -0.00369 | | std | 1.31 | | value_loss | 0.00888 | ------------------------------------------ -------------------------------- | time/ | | | fps | 1676 | | iterations | 223 | | time_elapsed | 2179 | | total_timesteps | 3653632 | -------------------------------- ----------------------------------------- | time/ | | | fps | 1679 | | iterations | 224 | | time_elapsed | 2185 | | total_timesteps | 3670016 | | train/ | | | approx_kl | 0.007888832 | | clip_fraction | 0.0783 | | clip_range | 0.2 | | entropy_loss | -3.37 | | explained_variance | 0.914 | | learning_rate | 0.0003 | | loss | -0.0298 | | n_updates | 2230 | | policy_gradient_loss | -0.00449 | | std | 1.32 | | value_loss | 0.00867 | ----------------------------------------- ------------------------------------------ | time/ | | | fps | 1682 | | iterations | 225 | | time_elapsed | 2190 | | total_timesteps | 3686400 | | train/ | | | approx_kl | 0.0069514583 | | clip_fraction | 0.0791 | | clip_range | 0.2 | | entropy_loss | -3.38 | | explained_variance | 0.946 | | learning_rate | 0.0003 | | loss | -0.0283 | | n_updates | 2240 | | policy_gradient_loss | -0.00427 | | std | 1.32 | | value_loss | 0.00382 | ------------------------------------------ Eval num_timesteps=3700000, episode_reward=19.29 +/- 50.45 Episode length: 2000.00 +/- 0.00 ----------------------------------------- | eval/ | | | mean_ep_length | 2e+03 | | mean_reward | 19.3 | | time/ | | | total_timesteps | 3700000 | | train/ | | | approx_kl | 0.008142319 | | clip_fraction | 0.0865 | | clip_range | 0.2 | | entropy_loss | -3.37 | | explained_variance | 0.92 | | learning_rate | 0.0003 | | loss | -0.0467 | | n_updates | 2250 | | policy_gradient_loss | -0.00506 | | std | 1.31 | | value_loss | 0.00547 | ----------------------------------------- -------------------------------- | time/ | | | fps | 1674 | | iterations | 226 | | time_elapsed | 2210 | | total_timesteps | 3702784 | -------------------------------- ------------------------------------------ | time/ | | | fps | 1677 | | iterations | 227 | | time_elapsed | 2216 | | total_timesteps | 3719168 | | train/ | | | approx_kl | 0.0077144434 | | clip_fraction | 0.0783 | | clip_range | 0.2 | | entropy_loss | -3.36 | | explained_variance | 0.931 | | learning_rate | 0.0003 | | loss | -0.0331 | | n_updates | 2260 | | policy_gradient_loss | -0.00529 | | std | 1.31 | | value_loss | 0.00486 | ------------------------------------------ ----------------------------------------- | time/ | | | fps | 1667 | | iterations | 228 | | time_elapsed | 2239 | | total_timesteps | 3735552 | | train/ | | | approx_kl | 0.007820845 | | clip_fraction | 0.087 | | clip_range | 0.2 | | entropy_loss | -3.37 | | explained_variance | 0.95 | | learning_rate | 0.0003 | | loss | -0.0321 | | n_updates | 2270 | | policy_gradient_loss | -0.00493 | | std | 1.31 | | value_loss | 0.00531 | ----------------------------------------- Eval num_timesteps=3750000, episode_reward=35.91 +/- 47.57 Episode length: 2000.00 +/- 0.00 ----------------------------------------- | eval/ | | | mean_ep_length | 2e+03 | | mean_reward | 35.9 | | time/ | | | total_timesteps | 3750000 | | train/ | | | approx_kl | 0.008380983 | | clip_fraction | 0.0868 | | clip_range | 0.2 | | entropy_loss | -3.37 | | explained_variance | 0.927 | | learning_rate | 0.0003 | | loss | -0.0318 | | n_updates | 2280 | | policy_gradient_loss | -0.0046 | | std | 1.32 | | value_loss | 0.00684 | ----------------------------------------- -------------------------------- | time/ | | | fps | 1639 | | iterations | 229 | | time_elapsed | 2289 | | total_timesteps | 3751936 | -------------------------------- ----------------------------------------- | time/ | | | fps | 1642 | | iterations | 230 | | time_elapsed | 2294 | | total_timesteps | 3768320 | | train/ | | | approx_kl | 0.007415652 | | clip_fraction | 0.0758 | | clip_range | 0.2 | | entropy_loss | -3.37 | | explained_variance | 0.953 | | learning_rate | 0.0003 | | loss | -0.0354 | | n_updates | 2290 | | policy_gradient_loss | -0.00557 | | std | 1.31 | | value_loss | 0.0122 | ----------------------------------------- ------------------------------------------ | time/ | | | fps | 1646 | | iterations | 231 | | time_elapsed | 2299 | | total_timesteps | 3784704 | | train/ | | | approx_kl | 0.0071868873 | | clip_fraction | 0.0736 | | clip_range | 0.2 | | entropy_loss | -3.38 | | explained_variance | 0.954 | | learning_rate | 0.0003 | | loss | -0.0457 | | n_updates | 2300 | | policy_gradient_loss | -0.00442 | | std | 1.33 | | value_loss | 0.0201 | ------------------------------------------ Eval num_timesteps=3800000, episode_reward=31.58 +/- 50.62 Episode length: 2000.00 +/- 0.00 ------------------------------------------ | eval/ | | | mean_ep_length | 2e+03 | | mean_reward | 31.6 | | time/ | | | total_timesteps | 3800000 | | train/ | | | approx_kl | 0.0074889637 | | clip_fraction | 0.0805 | | clip_range | 0.2 | | entropy_loss | -3.4 | | explained_variance | 0.95 | | learning_rate | 0.0003 | | loss | -0.0355 | | n_updates | 2310 | | policy_gradient_loss | -0.00474 | | std | 1.33 | | value_loss | 0.00892 | ------------------------------------------ [Diag @ 3,800,000 | n_sheep=5 | success=0%] NEVER_COMPACT 19/20 COMPACT_CANT_DRIVE 1/20 action_mag mean=0.128 p10=0.005 p90=0.475 (0=stopped, 1=full speed) min_flock_radius mean=8.35m best=4.80m (target <5m to compact) min_dog_to_com mean=0.71m best=0.23m (FLEE_DIST=7m) min_com_to_pen mean=13.72m best=8.54m reward/step (mean): progress=+0.0063 alignment=+0.0388 pen_bonus=+0.0010 step_cost=-0.0200 complete=+0.0000 -------------------------------- | time/ | | | fps | 1633 | | iterations | 232 | | time_elapsed | 2326 | | total_timesteps | 3801088 | -------------------------------- ------------------------------------------ | time/ | | | fps | 1636 | | iterations | 233 | | time_elapsed | 2332 | | total_timesteps | 3817472 | | train/ | | | approx_kl | 0.0070604184 | | clip_fraction | 0.0765 | | clip_range | 0.2 | | entropy_loss | -3.4 | | explained_variance | 0.953 | | learning_rate | 0.0003 | | loss | -0.0398 | | n_updates | 2320 | | policy_gradient_loss | -0.00453 | | std | 1.33 | | value_loss | 0.00675 | ------------------------------------------ ----------------------------------------- | time/ | | | fps | 1640 | | iterations | 234 | | time_elapsed | 2336 | | total_timesteps | 3833856 | | train/ | | | approx_kl | 0.007709453 | | clip_fraction | 0.0816 | | clip_range | 0.2 | | entropy_loss | -3.4 | | explained_variance | 0.943 | | learning_rate | 0.0003 | | loss | -0.0359 | | n_updates | 2330 | | policy_gradient_loss | -0.00423 | | std | 1.34 | | value_loss | 0.00754 | ----------------------------------------- Eval num_timesteps=3850000, episode_reward=42.98 +/- 33.36 Episode length: 2000.00 +/- 0.00 ----------------------------------------- | eval/ | | | mean_ep_length | 2e+03 | | mean_reward | 43 | | time/ | | | total_timesteps | 3850000 | | train/ | | | approx_kl | 0.007679659 | | clip_fraction | 0.0858 | | clip_range | 0.2 | | entropy_loss | -3.4 | | explained_variance | 0.961 | | learning_rate | 0.0003 | | loss | -0.032 | | n_updates | 2340 | | policy_gradient_loss | -0.00716 | | std | 1.33 | | value_loss | 0.00907 | ----------------------------------------- -------------------------------- | time/ | | | fps | 1635 | | iterations | 235 | | time_elapsed | 2354 | | total_timesteps | 3850240 | -------------------------------- ------------------------------------------ | time/ | | | fps | 1638 | | iterations | 236 | | time_elapsed | 2360 | | total_timesteps | 3866624 | | train/ | | | approx_kl | 0.0077598644 | | clip_fraction | 0.0848 | | clip_range | 0.2 | | entropy_loss | -3.38 | | explained_variance | 0.96 | | learning_rate | 0.0003 | | loss | -0.0468 | | n_updates | 2350 | | policy_gradient_loss | -0.005 | | std | 1.33 | | value_loss | 0.0101 | ------------------------------------------ ------------------------------------------ | time/ | | | fps | 1641 | | iterations | 237 | | time_elapsed | 2366 | | total_timesteps | 3883008 | | train/ | | | approx_kl | 0.0068941545 | | clip_fraction | 0.0673 | | clip_range | 0.2 | | entropy_loss | -3.39 | | explained_variance | 0.96 | | learning_rate | 0.0003 | | loss | -0.0398 | | n_updates | 2360 | | policy_gradient_loss | -0.0047 | | std | 1.33 | | value_loss | 0.0113 | ------------------------------------------ ------------------------------------------ | time/ | | | fps | 1643 | | iterations | 238 | | time_elapsed | 2372 | | total_timesteps | 3899392 | | train/ | | | approx_kl | 0.0073663425 | | clip_fraction | 0.0785 | | clip_range | 0.2 | | entropy_loss | -3.41 | | explained_variance | 0.963 | | learning_rate | 0.0003 | | loss | -0.0319 | | n_updates | 2370 | | policy_gradient_loss | -0.00458 | | std | 1.35 | | value_loss | 0.0036 | ------------------------------------------ Eval num_timesteps=3900000, episode_reward=33.74 +/- 40.96 Episode length: 2000.00 +/- 0.00 ----------------------------------------- | eval/ | | | mean_ep_length | 2e+03 | | mean_reward | 33.7 | | time/ | | | total_timesteps | 3900000 | | train/ | | | approx_kl | 0.007122398 | | clip_fraction | 0.0759 | | clip_range | 0.2 | | entropy_loss | -3.41 | | explained_variance | 0.972 | | learning_rate | 0.0003 | | loss | -0.0383 | | n_updates | 2380 | | policy_gradient_loss | -0.00446 | | std | 1.35 | | value_loss | 0.00445 | ----------------------------------------- -------------------------------- | time/ | | | fps | 1637 | | iterations | 239 | | time_elapsed | 2391 | | total_timesteps | 3915776 | -------------------------------- ----------------------------------------- | time/ | | | fps | 1640 | | iterations | 240 | | time_elapsed | 2396 | | total_timesteps | 3932160 | | train/ | | | approx_kl | 0.008265208 | | clip_fraction | 0.0845 | | clip_range | 0.2 | | entropy_loss | -3.41 | | explained_variance | 0.926 | | learning_rate | 0.0003 | | loss | -0.0361 | | n_updates | 2390 | | policy_gradient_loss | -0.00536 | | std | 1.34 | | value_loss | 0.00846 | ----------------------------------------- ----------------------------------------- | time/ | | | fps | 1643 | | iterations | 241 | | time_elapsed | 2402 | | total_timesteps | 3948544 | | train/ | | | approx_kl | 0.008583728 | | clip_fraction | 0.0893 | | clip_range | 0.2 | | entropy_loss | -3.42 | | explained_variance | 0.915 | | learning_rate | 0.0003 | | loss | -0.0297 | | n_updates | 2400 | | policy_gradient_loss | -0.00592 | | std | 1.35 | | value_loss | 0.0068 | ----------------------------------------- Eval num_timesteps=3950000, episode_reward=46.06 +/- 34.67 Episode length: 2000.00 +/- 0.00 ------------------------------------------ | eval/ | | | mean_ep_length | 2e+03 | | mean_reward | 46.1 | | time/ | | | total_timesteps | 3950000 | | train/ | | | approx_kl | 0.0060660206 | | clip_fraction | 0.0654 | | clip_range | 0.2 | | entropy_loss | -3.42 | | explained_variance | 0.942 | | learning_rate | 0.0003 | | loss | -0.0359 | | n_updates | 2410 | | policy_gradient_loss | -0.0038 | | std | 1.35 | | value_loss | 0.00296 | ------------------------------------------ -------------------------------- | time/ | | | fps | 1637 | | iterations | 242 | | time_elapsed | 2421 | | total_timesteps | 3964928 | -------------------------------- ----------------------------------------- | time/ | | | fps | 1639 | | iterations | 243 | | time_elapsed | 2427 | | total_timesteps | 3981312 | | train/ | | | approx_kl | 0.007591601 | | clip_fraction | 0.0808 | | clip_range | 0.2 | | entropy_loss | -3.4 | | explained_variance | 0.964 | | learning_rate | 0.0003 | | loss | -0.0386 | | n_updates | 2420 | | policy_gradient_loss | -0.00575 | | std | 1.34 | | value_loss | 0.00714 | ----------------------------------------- ----------------------------------------- | time/ | | | fps | 1642 | | iterations | 244 | | time_elapsed | 2433 | | total_timesteps | 3997696 | | train/ | | | approx_kl | 0.006255053 | | clip_fraction | 0.0663 | | clip_range | 0.2 | | entropy_loss | -3.41 | | explained_variance | 0.939 | | learning_rate | 0.0003 | | loss | -0.0304 | | n_updates | 2430 | | policy_gradient_loss | -0.00497 | | std | 1.35 | | value_loss | 0.00585 | ----------------------------------------- Eval num_timesteps=4000000, episode_reward=19.52 +/- 38.43 Episode length: 2000.00 +/- 0.00 ----------------------------------------- | eval/ | | | mean_ep_length | 2e+03 | | mean_reward | 19.5 | | time/ | | | total_timesteps | 4000000 | | train/ | | | approx_kl | 0.008279499 | | clip_fraction | 0.0814 | | clip_range | 0.2 | | entropy_loss | -3.4 | | explained_variance | 0.958 | | learning_rate | 0.0003 | | loss | -0.0379 | | n_updates | 2440 | | policy_gradient_loss | -0.00568 | | std | 1.34 | | value_loss | 0.00469 | ----------------------------------------- [Diag @ 4,000,000 | n_sheep=5 | success=0%] NEVER_COMPACT 20/20 action_mag mean=0.158 p10=0.006 p90=0.744 (0=stopped, 1=full speed) min_flock_radius mean=8.94m best=6.34m (target <5m to compact) min_dog_to_com mean=0.82m best=0.49m (FLEE_DIST=7m) min_com_to_pen mean=13.86m best=7.80m reward/step (mean): progress=+0.0029 alignment=+0.0397 pen_bonus=+0.0003 step_cost=-0.0200 complete=+0.0000 -------------------------------- | time/ | | | fps | 1630 | | iterations | 245 | | time_elapsed | 2461 | | total_timesteps | 4014080 | -------------------------------- Training complete. Artefacts saved to runs/ppo_debug/