Using cpu device Logging to runs/ppo_fix_check2/ppo_1 ------------------------------ | time/ | | | fps | 4605 | | iterations | 1 | | time_elapsed | 3 | | total_timesteps | 16384 | ------------------------------ ------------------------------------------ | time/ | | | fps | 4011 | | iterations | 2 | | time_elapsed | 8 | | total_timesteps | 32768 | | train/ | | | approx_kl | 0.0033352287 | | clip_fraction | 0.0253 | | clip_range | 0.2 | | entropy_loss | -2.83 | | explained_variance | 0.271 | | learning_rate | 0.0003 | | loss | -0.00687 | | n_updates | 10 | | policy_gradient_loss | -0.00103 | | std | 0.996 | | value_loss | 0.0684 | ------------------------------------------ ----------------------------------------- | time/ | | | fps | 3789 | | iterations | 3 | | time_elapsed | 12 | | total_timesteps | 49152 | | train/ | | | approx_kl | 0.005950423 | | clip_fraction | 0.0552 | | clip_range | 0.2 | | entropy_loss | -2.83 | | explained_variance | 0.527 | | learning_rate | 0.0003 | | loss | -0.0153 | | n_updates | 20 | | policy_gradient_loss | -0.0029 | | std | 0.997 | | value_loss | 0.0663 | ----------------------------------------- /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=-25.68 +/- 59.67 Episode length: 1815.95 +/- 456.88 ------------------------------------------ | eval/ | | | mean_ep_length | 1.82e+03 | | mean_reward | -25.7 | | time/ | | | total_timesteps | 50000 | | train/ | | | approx_kl | 0.0040030424 | | clip_fraction | 0.0356 | | clip_range | 0.2 | | entropy_loss | -2.85 | | explained_variance | 0.421 | | learning_rate | 0.0003 | | loss | 0.149 | | n_updates | 30 | | policy_gradient_loss | -0.00198 | | std | 1.01 | | value_loss | 0.114 | ------------------------------------------ New best mean reward! ------------------------------ | time/ | | | fps | 2351 | | iterations | 4 | | time_elapsed | 27 | | total_timesteps | 65536 | ------------------------------ ----------------------------------------- | time/ | | | fps | 2446 | | iterations | 5 | | time_elapsed | 33 | | total_timesteps | 81920 | | train/ | | | approx_kl | 0.005522004 | | clip_fraction | 0.0604 | | clip_range | 0.2 | | entropy_loss | -2.86 | | explained_variance | 0.737 | | learning_rate | 0.0003 | | loss | -0.0301 | | n_updates | 40 | | policy_gradient_loss | -0.00434 | | std | 1.01 | | value_loss | 0.0164 | ----------------------------------------- ------------------------------------------ | time/ | | | fps | 2617 | | iterations | 6 | | time_elapsed | 37 | | total_timesteps | 98304 | | train/ | | | approx_kl | 0.0052388343 | | clip_fraction | 0.0463 | | clip_range | 0.2 | | entropy_loss | -2.86 | | explained_variance | 0.626 | | learning_rate | 0.0003 | | loss | -0.0294 | | n_updates | 50 | | policy_gradient_loss | -0.00297 | | std | 1.01 | | value_loss | 0.0597 | ------------------------------------------ /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=-22.76 +/- 46.60 Episode length: 1900.95 +/- 430.60 ----------------------------------------- | eval/ | | | mean_ep_length | 1.9e+03 | | mean_reward | -22.8 | | time/ | | | total_timesteps | 100000 | | train/ | | | approx_kl | 0.005612197 | | clip_fraction | 0.0475 | | clip_range | 0.2 | | entropy_loss | -2.86 | | explained_variance | 0.747 | | learning_rate | 0.0003 | | loss | -0.0261 | | n_updates | 60 | | policy_gradient_loss | -0.00393 | | std | 1.01 | | value_loss | 0.0517 | ----------------------------------------- New best mean reward! ------------------------------- | time/ | | | fps | 2178 | | iterations | 7 | | time_elapsed | 52 | | total_timesteps | 114688 | ------------------------------- ------------------------------------------ | time/ | | | fps | 2294 | | iterations | 8 | | time_elapsed | 57 | | total_timesteps | 131072 | | train/ | | | approx_kl | 0.0057119504 | | clip_fraction | 0.0541 | | clip_range | 0.2 | | entropy_loss | -2.85 | | explained_variance | 0.896 | | learning_rate | 0.0003 | | loss | -0.0144 | | n_updates | 70 | | policy_gradient_loss | -0.00364 | | std | 1 | | value_loss | 0.0738 | ------------------------------------------ ----------------------------------------- | time/ | | | fps | 2393 | | iterations | 9 | | time_elapsed | 61 | | total_timesteps | 147456 | | train/ | | | approx_kl | 0.005940904 | | clip_fraction | 0.0565 | | clip_range | 0.2 | | entropy_loss | -2.85 | | explained_variance | 0.89 | | learning_rate | 0.0003 | | loss | -0.0283 | | n_updates | 80 | | policy_gradient_loss | -0.00245 | | std | 1.01 | | value_loss | 0.0761 | ----------------------------------------- Eval num_timesteps=150000, episode_reward=-29.37 +/- 28.32 Episode length: 1997.50 +/- 10.90 ----------------------------------------- | eval/ | | | mean_ep_length | 2e+03 | | mean_reward | -29.4 | | time/ | | | total_timesteps | 150000 | | train/ | | | approx_kl | 0.004531667 | | clip_fraction | 0.0392 | | clip_range | 0.2 | | entropy_loss | -2.85 | | explained_variance | 0.958 | | learning_rate | 0.0003 | | loss | -0.0343 | | n_updates | 90 | | policy_gradient_loss | -0.00379 | | std | 1.01 | | value_loss | 0.00995 | ----------------------------------------- [Diag @ 150,000 | n_sheep=1 | success=0%] COMPACT_CANT_DRIVE 17/20 DROVE_NO_SHEEP 3/20 action_mag mean=0.089 p10=0.003 p90=0.274 (0=stopped, 1=full speed) min_flock_radius mean=0.00m best=0.00m (target <5m to compact) min_dog_to_com mean=4.40m best=2.07m (FLEE_DIST=7m) min_com_to_pen mean=11.66m best=1.50m reward/step (mean): progress=+0.0004 alignment=+0.0000 pen_bonus=+0.0000 step_cost=-0.0200 complete=+0.0000 ------------------------------- | time/ | | | fps | 1950 | | iterations | 10 | | time_elapsed | 84 | | total_timesteps | 163840 | ------------------------------- ------------------------------------------ | time/ | | | fps | 2020 | | iterations | 11 | | time_elapsed | 89 | | total_timesteps | 180224 | | train/ | | | approx_kl | 0.0061831754 | | clip_fraction | 0.068 | | clip_range | 0.2 | | entropy_loss | -2.86 | | explained_variance | 0.975 | | learning_rate | 0.0003 | | loss | -0.0349 | | n_updates | 100 | | policy_gradient_loss | -0.00607 | | std | 1.02 | | value_loss | 0.0156 | ------------------------------------------ ----------------------------------------- | time/ | | | fps | 2084 | | iterations | 12 | | time_elapsed | 94 | | total_timesteps | 196608 | | train/ | | | approx_kl | 0.009407628 | | clip_fraction | 0.123 | | clip_range | 0.2 | | entropy_loss | -2.87 | | explained_variance | 0.899 | | learning_rate | 0.0003 | | loss | -0.0305 | | n_updates | 110 | | policy_gradient_loss | -0.00932 | | std | 1.02 | | value_loss | 0.0223 | ----------------------------------------- Eval num_timesteps=200000, episode_reward=-12.36 +/- 51.37 Episode length: 1880.20 +/- 355.04 ----------------------------------------- | eval/ | | | mean_ep_length | 1.88e+03 | | mean_reward | -12.4 | | time/ | | | total_timesteps | 200000 | | train/ | | | approx_kl | 0.008270489 | | clip_fraction | 0.0945 | | clip_range | 0.2 | | entropy_loss | -2.85 | | explained_variance | 0.945 | | learning_rate | 0.0003 | | loss | -0.0339 | | n_updates | 120 | | policy_gradient_loss | -0.00809 | | std | 1 | | value_loss | 0.0162 | ----------------------------------------- New best mean reward! ------------------------------- | time/ | | | fps | 1936 | | iterations | 13 | | time_elapsed | 109 | | total_timesteps | 212992 | ------------------------------- ----------------------------------------- | time/ | | | fps | 1989 | | iterations | 14 | | time_elapsed | 115 | | total_timesteps | 229376 | | train/ | | | approx_kl | 0.008541125 | | clip_fraction | 0.112 | | clip_range | 0.2 | | entropy_loss | -2.83 | | explained_variance | 0.944 | | learning_rate | 0.0003 | | loss | -0.0184 | | n_updates | 130 | | policy_gradient_loss | -0.00846 | | std | 0.994 | | value_loss | 0.0284 | ----------------------------------------- ---------------------------------------- | time/ | | | fps | 2037 | | iterations | 15 | | time_elapsed | 120 | | total_timesteps | 245760 | | train/ | | | approx_kl | 0.00763176 | | clip_fraction | 0.0894 | | clip_range | 0.2 | | entropy_loss | -2.81 | | explained_variance | 0.9 | | learning_rate | 0.0003 | | loss | -0.0128 | | n_updates | 140 | | policy_gradient_loss | -0.00655 | | std | 0.987 | | value_loss | 0.071 | ---------------------------------------- Eval num_timesteps=250000, episode_reward=45.82 +/- 68.33 Episode length: 1391.70 +/- 757.58 ----------------------------------------- | eval/ | | | mean_ep_length | 1.39e+03 | | mean_reward | 45.8 | | time/ | | | total_timesteps | 250000 | | train/ | | | approx_kl | 0.009210973 | | clip_fraction | 0.11 | | clip_range | 0.2 | | entropy_loss | -2.81 | | explained_variance | 0.95 | | learning_rate | 0.0003 | | loss | -0.0401 | | n_updates | 150 | | policy_gradient_loss | -0.0082 | | std | 0.986 | | value_loss | 0.0202 | ----------------------------------------- New best mean reward! ------------------------------- | time/ | | | fps | 1958 | | iterations | 16 | | time_elapsed | 133 | | total_timesteps | 262144 | ------------------------------- ----------------------------------------- | time/ | | | fps | 2005 | | iterations | 17 | | time_elapsed | 138 | | total_timesteps | 278528 | | train/ | | | approx_kl | 0.008197077 | | clip_fraction | 0.096 | | clip_range | 0.2 | | entropy_loss | -2.79 | | explained_variance | 0.949 | | learning_rate | 0.0003 | | loss | -0.0375 | | n_updates | 160 | | policy_gradient_loss | -0.00834 | | std | 0.976 | | value_loss | 0.0207 | ----------------------------------------- ----------------------------------------- | time/ | | | fps | 2061 | | iterations | 18 | | time_elapsed | 143 | | total_timesteps | 294912 | | train/ | | | approx_kl | 0.006078005 | | clip_fraction | 0.0598 | | clip_range | 0.2 | | entropy_loss | -2.78 | | explained_variance | 0.965 | | learning_rate | 0.0003 | | loss | -0.0188 | | n_updates | 170 | | policy_gradient_loss | -0.00464 | | std | 0.969 | | value_loss | 0.0178 | ----------------------------------------- Eval num_timesteps=300000, episode_reward=56.19 +/- 63.26 Episode length: 1246.75 +/- 843.82 ------------------------------------------ | eval/ | | | mean_ep_length | 1.25e+03 | | mean_reward | 56.2 | | time/ | | | total_timesteps | 300000 | | train/ | | | approx_kl | 0.0056289425 | | clip_fraction | 0.0523 | | clip_range | 0.2 | | entropy_loss | -2.76 | | explained_variance | 0.969 | | learning_rate | 0.0003 | | loss | -0.0246 | | n_updates | 180 | | policy_gradient_loss | -0.00378 | | std | 0.961 | | value_loss | 0.0174 | ------------------------------------------ New best mean reward! [Diag @ 300,000 | n_sheep=1 | success=40%] DROVE_NO_SHEEP 11/20 SUCCESS 8/20 COMPACT_CANT_DRIVE 1/20 action_mag mean=0.076 p10=0.000 p90=0.193 (0=stopped, 1=full speed) min_flock_radius mean=0.00m best=0.00m (target <5m to compact) min_dog_to_com mean=2.83m best=0.24m (FLEE_DIST=7m) min_com_to_pen mean=2.99m best=1.50m reward/step (mean): progress=+0.0236 alignment=+0.0012 pen_bonus=+0.0029 step_cost=-0.0200 complete=+0.0291 ------------------------------- | time/ | | | fps | 1939 | | iterations | 19 | | time_elapsed | 160 | | total_timesteps | 311296 | ------------------------------- ----------------------------------------- | time/ | | | fps | 1983 | | iterations | 20 | | time_elapsed | 165 | | total_timesteps | 327680 | | train/ | | | approx_kl | 0.005042998 | | clip_fraction | 0.05 | | clip_range | 0.2 | | entropy_loss | -2.73 | | explained_variance | 0.941 | | learning_rate | 0.0003 | | loss | -0.0242 | | n_updates | 190 | | policy_gradient_loss | -0.00399 | | std | 0.947 | | value_loss | 0.00505 | ----------------------------------------- ------------------------------------------ | time/ | | | fps | 2018 | | iterations | 21 | | time_elapsed | 170 | | total_timesteps | 344064 | | train/ | | | approx_kl | 0.0054986854 | | clip_fraction | 0.0569 | | clip_range | 0.2 | | entropy_loss | -2.72 | | explained_variance | 0.942 | | learning_rate | 0.0003 | | loss | -0.0248 | | n_updates | 200 | | policy_gradient_loss | -0.00415 | | std | 0.941 | | value_loss | 0.00784 | ------------------------------------------ Eval num_timesteps=350000, episode_reward=25.08 +/- 61.55 Episode length: 1562.00 +/- 761.23 ------------------------------------------ | eval/ | | | mean_ep_length | 1.56e+03 | | mean_reward | 25.1 | | time/ | | | total_timesteps | 350000 | | train/ | | | approx_kl | 0.0046333643 | | clip_fraction | 0.0476 | | clip_range | 0.2 | | entropy_loss | -2.71 | | explained_variance | 0.934 | | learning_rate | 0.0003 | | loss | -0.0244 | | n_updates | 210 | | policy_gradient_loss | -0.00237 | | std | 0.934 | | value_loss | 0.00827 | ------------------------------------------ ------------------------------- | time/ | | | fps | 1950 | | iterations | 22 | | time_elapsed | 184 | | total_timesteps | 360448 | ------------------------------- ----------------------------------------- | time/ | | | fps | 1990 | | iterations | 23 | | time_elapsed | 189 | | total_timesteps | 376832 | | train/ | | | approx_kl | 0.006686668 | | clip_fraction | 0.0757 | | clip_range | 0.2 | | entropy_loss | -2.7 | | explained_variance | 0.963 | | learning_rate | 0.0003 | | loss | -0.0423 | | n_updates | 220 | | policy_gradient_loss | -0.00244 | | std | 0.936 | | value_loss | 0.00575 | ----------------------------------------- ----------------------------------------- | time/ | | | fps | 2027 | | iterations | 24 | | time_elapsed | 193 | | total_timesteps | 393216 | | train/ | | | approx_kl | 0.009116547 | | clip_fraction | 0.103 | | clip_range | 0.2 | | entropy_loss | -2.71 | | explained_variance | 0.97 | | learning_rate | 0.0003 | | loss | -0.0353 | | n_updates | 230 | | policy_gradient_loss | -0.0042 | | std | 0.941 | | value_loss | 0.006 | ----------------------------------------- Eval num_timesteps=400000, episode_reward=56.91 +/- 71.91 Episode length: 1225.25 +/- 861.21 ------------------------------------------ | eval/ | | | mean_ep_length | 1.23e+03 | | mean_reward | 56.9 | | time/ | | | total_timesteps | 400000 | | train/ | | | approx_kl | 0.0061917743 | | clip_fraction | 0.0658 | | clip_range | 0.2 | | entropy_loss | -2.72 | | explained_variance | 0.975 | | learning_rate | 0.0003 | | loss | -0.0378 | | n_updates | 240 | | policy_gradient_loss | -0.00282 | | std | 0.943 | | value_loss | 0.00633 | ------------------------------------------ New best mean reward! ------------------------------- | time/ | | | fps | 1981 | | iterations | 25 | | time_elapsed | 206 | | total_timesteps | 409600 | ------------------------------- ----------------------------------------- | time/ | | | fps | 2011 | | iterations | 26 | | time_elapsed | 211 | | total_timesteps | 425984 | | train/ | | | approx_kl | 0.007945089 | | clip_fraction | 0.1 | | clip_range | 0.2 | | entropy_loss | -2.73 | | explained_variance | 0.978 | | learning_rate | 0.0003 | | loss | -0.0343 | | n_updates | 250 | | policy_gradient_loss | -0.00475 | | std | 0.95 | | value_loss | 0.00708 | ----------------------------------------- ----------------------------------------- | time/ | | | fps | 2044 | | iterations | 27 | | time_elapsed | 216 | | total_timesteps | 442368 | | train/ | | | approx_kl | 0.013059773 | | clip_fraction | 0.152 | | clip_range | 0.2 | | entropy_loss | -2.76 | | explained_variance | 0.984 | | learning_rate | 0.0003 | | loss | -0.0421 | | n_updates | 260 | | policy_gradient_loss | -0.00542 | | std | 0.967 | | value_loss | 0.00331 | ----------------------------------------- Eval num_timesteps=450000, episode_reward=58.80 +/- 74.46 Episode length: 1123.15 +/- 881.85 ------------------------------------------ | eval/ | | | mean_ep_length | 1.12e+03 | | mean_reward | 58.8 | | time/ | | | total_timesteps | 450000 | | train/ | | | approx_kl | 0.0085322345 | | clip_fraction | 0.0967 | | clip_range | 0.2 | | entropy_loss | -2.77 | | explained_variance | 0.98 | | learning_rate | 0.0003 | | loss | -0.0264 | | n_updates | 270 | | policy_gradient_loss | -0.00612 | | std | 0.963 | | value_loss | 0.00919 | ------------------------------------------ New best mean reward! [Diag @ 450,000 | n_sheep=1 | success=65%] SUCCESS 13/20 DROVE_NO_SHEEP 4/20 COMPACT_CANT_DRIVE 3/20 action_mag mean=0.105 p10=0.000 p90=0.272 (0=stopped, 1=full speed) min_flock_radius mean=0.00m best=0.00m (target <5m to compact) min_dog_to_com mean=1.67m best=0.43m (FLEE_DIST=7m) min_com_to_pen mean=3.26m best=2.29m reward/step (mean): progress=+0.0326 alignment=+0.0024 pen_bonus=+0.0076 step_cost=-0.0200 complete=+0.0762 ------------------------------- | time/ | | | fps | 1974 | | iterations | 28 | | time_elapsed | 232 | | total_timesteps | 458752 | ------------------------------- ---------------------------------------- | time/ | | | fps | 2005 | | iterations | 29 | | time_elapsed | 236 | | total_timesteps | 475136 | | train/ | | | approx_kl | 0.01203198 | | clip_fraction | 0.146 | | clip_range | 0.2 | | entropy_loss | -2.79 | | explained_variance | 0.963 | | learning_rate | 0.0003 | | loss | 0.00738 | | n_updates | 280 | | policy_gradient_loss | -0.0128 | | std | 0.982 | | value_loss | 0.0749 | ---------------------------------------- ------------------------------------------ | time/ | | | fps | 2037 | | iterations | 30 | | time_elapsed | 241 | | total_timesteps | 491520 | | train/ | | | approx_kl | 0.0078244675 | | clip_fraction | 0.0856 | | clip_range | 0.2 | | entropy_loss | -2.8 | | explained_variance | 0.937 | | learning_rate | 0.0003 | | loss | 0.0631 | | n_updates | 290 | | policy_gradient_loss | -0.00651 | | std | 0.977 | | value_loss | 0.131 | ------------------------------------------ Eval num_timesteps=500000, episode_reward=135.29 +/- 9.81 Episode length: 287.30 +/- 88.71 ---------------------------------------- | eval/ | | | mean_ep_length | 287 | | mean_reward | 135 | | time/ | | | total_timesteps | 500000 | | train/ | | | approx_kl | 0.00837522 | | clip_fraction | 0.0866 | | clip_range | 0.2 | | entropy_loss | -2.77 | | explained_variance | 0.948 | | learning_rate | 0.0003 | | loss | 0.041 | | n_updates | 300 | | policy_gradient_loss | -0.00532 | | std | 0.962 | | value_loss | 0.0898 | ---------------------------------------- New best mean reward! ------------------------------- | time/ | | | fps | 2048 | | iterations | 31 | | time_elapsed | 247 | | total_timesteps | 507904 | ------------------------------- ------------------------------------------ | time/ | | | fps | 2070 | | iterations | 32 | | time_elapsed | 253 | | total_timesteps | 524288 | | train/ | | | approx_kl | 0.0067581255 | | clip_fraction | 0.0543 | | clip_range | 0.2 | | entropy_loss | -2.75 | | explained_variance | 0.932 | | learning_rate | 0.0003 | | loss | 0.0518 | | n_updates | 310 | | policy_gradient_loss | -0.00297 | | std | 0.954 | | value_loss | 0.111 | ------------------------------------------ ------------------------------------------ | time/ | | | fps | 2090 | | iterations | 33 | | time_elapsed | 258 | | total_timesteps | 540672 | | train/ | | | approx_kl | 0.0066835573 | | clip_fraction | 0.0597 | | clip_range | 0.2 | | entropy_loss | -2.74 | | explained_variance | 0.934 | | learning_rate | 0.0003 | | loss | 0.00545 | | n_updates | 320 | | policy_gradient_loss | -0.00508 | | std | 0.949 | | value_loss | 0.101 | ------------------------------------------ Eval num_timesteps=550000, episode_reward=136.08 +/- 11.93 Episode length: 285.80 +/- 123.59 ------------------------------------------ | eval/ | | | mean_ep_length | 286 | | mean_reward | 136 | | time/ | | | total_timesteps | 550000 | | train/ | | | approx_kl | 0.0062076193 | | clip_fraction | 0.0672 | | clip_range | 0.2 | | entropy_loss | -2.71 | | explained_variance | 0.942 | | learning_rate | 0.0003 | | loss | 0.0229 | | n_updates | 330 | | policy_gradient_loss | -0.00616 | | std | 0.933 | | value_loss | 0.0813 | ------------------------------------------ New best mean reward! ------------------------------- | time/ | | | fps | 2104 | | iterations | 34 | | time_elapsed | 264 | | total_timesteps | 557056 | ------------------------------- ------------------------------------------ | time/ | | | fps | 2130 | | iterations | 35 | | time_elapsed | 269 | | total_timesteps | 573440 | | train/ | | | approx_kl | 0.0064913128 | | clip_fraction | 0.0631 | | clip_range | 0.2 | | entropy_loss | -2.67 | | explained_variance | 0.971 | | learning_rate | 0.0003 | | loss | -0.0199 | | n_updates | 340 | | policy_gradient_loss | -0.00631 | | std | 0.917 | | value_loss | 0.0185 | ------------------------------------------ ------------------------------------------ | time/ | | | fps | 2155 | | iterations | 36 | | time_elapsed | 273 | | total_timesteps | 589824 | | train/ | | | approx_kl | 0.0067110434 | | clip_fraction | 0.0719 | | clip_range | 0.2 | | entropy_loss | -2.63 | | explained_variance | 0.98 | | learning_rate | 0.0003 | | loss | -0.0343 | | n_updates | 350 | | policy_gradient_loss | -0.0069 | | std | 0.897 | | value_loss | 0.0113 | ------------------------------------------ Eval num_timesteps=600000, episode_reward=135.45 +/- 12.96 Episode length: 273.05 +/- 118.26 ------------------------------------------ | eval/ | | | mean_ep_length | 273 | | mean_reward | 135 | | time/ | | | total_timesteps | 600000 | | train/ | | | approx_kl | 0.0054842415 | | clip_fraction | 0.0564 | | clip_range | 0.2 | | entropy_loss | -2.59 | | explained_variance | 0.983 | | learning_rate | 0.0003 | | loss | -0.033 | | n_updates | 360 | | policy_gradient_loss | -0.0042 | | std | 0.883 | | value_loss | 0.00479 | ------------------------------------------ [Diag @ 600,000 | n_sheep=1 | success=100%] SUCCESS 20/20 action_mag mean=0.343 p10=0.232 p90=0.548 (0=stopped, 1=full speed) min_flock_radius mean=0.00m best=0.00m (target <5m to compact) min_dog_to_com mean=1.53m best=0.76m (FLEE_DIST=7m) min_com_to_pen mean=3.49m best=2.84m reward/step (mean): progress=+0.1066 alignment=+0.0088 pen_bonus=+0.0357 step_cost=-0.0200 complete=+0.3567 [Curriculum] leaving stage n_sheep=1 after 600,000 steps | training success rate (last 100 eps) = 100% [Curriculum] → 2 sheep at step 600,000 ------------------------------- | time/ | | | fps | 2156 | | iterations | 37 | | time_elapsed | 281 | | total_timesteps | 606208 | ------------------------------- ----------------------------------------- | time/ | | | fps | 2173 | | iterations | 38 | | time_elapsed | 286 | | total_timesteps | 622592 | | train/ | | | approx_kl | 0.011170821 | | clip_fraction | 0.117 | | clip_range | 0.2 | | entropy_loss | -2.59 | | explained_variance | 0.924 | | learning_rate | 0.0003 | | loss | -0.0137 | | n_updates | 370 | | policy_gradient_loss | 0.00714 | | std | 0.886 | | value_loss | 0.0417 | ----------------------------------------- ----------------------------------------- | time/ | | | fps | 2192 | | iterations | 39 | | time_elapsed | 291 | | total_timesteps | 638976 | | train/ | | | approx_kl | 0.012632904 | | clip_fraction | 0.156 | | clip_range | 0.2 | | entropy_loss | -2.6 | | explained_variance | 0.858 | | learning_rate | 0.0003 | | loss | -0.00445 | | n_updates | 380 | | policy_gradient_loss | 0.00112 | | std | 0.892 | | value_loss | 0.0156 | ----------------------------------------- Eval num_timesteps=650000, episode_reward=-38.36 +/- 29.94 Episode length: 2000.00 +/- 0.00 ----------------------------------------- | eval/ | | | mean_ep_length | 2e+03 | | mean_reward | -38.4 | | time/ | | | total_timesteps | 650000 | | train/ | | | approx_kl | 0.012015635 | | clip_fraction | 0.133 | | clip_range | 0.2 | | entropy_loss | -2.62 | | explained_variance | 0.946 | | learning_rate | 0.0003 | | loss | -0.0168 | | n_updates | 390 | | policy_gradient_loss | -0.000726 | | std | 0.904 | | value_loss | 0.0126 | ----------------------------------------- ------------------------------- | time/ | | | fps | 2131 | | iterations | 40 | | time_elapsed | 307 | | total_timesteps | 655360 | ------------------------------- ----------------------------------------- | time/ | | | fps | 2145 | | iterations | 41 | | time_elapsed | 313 | | total_timesteps | 671744 | | train/ | | | approx_kl | 0.009391339 | | clip_fraction | 0.121 | | clip_range | 0.2 | | entropy_loss | -2.63 | | explained_variance | 0.955 | | learning_rate | 0.0003 | | loss | -0.0164 | | n_updates | 400 | | policy_gradient_loss | -0.00177 | | std | 0.905 | | value_loss | 0.00536 | ----------------------------------------- ------------------------------------------ | time/ | | | fps | 2156 | | iterations | 42 | | time_elapsed | 319 | | total_timesteps | 688128 | | train/ | | | approx_kl | 0.0077482145 | | clip_fraction | 0.0977 | | clip_range | 0.2 | | entropy_loss | -2.64 | | explained_variance | 0.895 | | learning_rate | 0.0003 | | loss | -0.023 | | n_updates | 410 | | policy_gradient_loss | -0.00158 | | std | 0.908 | | value_loss | 0.0068 | ------------------------------------------ Eval num_timesteps=700000, episode_reward=-16.26 +/- 48.54 Episode length: 1934.20 +/- 286.82 ----------------------------------------- | eval/ | | | mean_ep_length | 1.93e+03 | | mean_reward | -16.3 | | time/ | | | total_timesteps | 700000 | | train/ | | | approx_kl | 0.007948186 | | clip_fraction | 0.0933 | | clip_range | 0.2 | | entropy_loss | -2.64 | | explained_variance | 0.934 | | learning_rate | 0.0003 | | loss | -0.0205 | | n_updates | 420 | | policy_gradient_loss | -0.00233 | | std | 0.904 | | value_loss | 0.00556 | ----------------------------------------- ------------------------------- | time/ | | | fps | 2093 | | iterations | 43 | | time_elapsed | 336 | | total_timesteps | 704512 | ------------------------------- ------------------------------------------ | time/ | | | fps | 2109 | | iterations | 44 | | time_elapsed | 341 | | total_timesteps | 720896 | | train/ | | | approx_kl | 0.0077707805 | | clip_fraction | 0.101 | | clip_range | 0.2 | | entropy_loss | -2.64 | | explained_variance | 0.929 | | learning_rate | 0.0003 | | loss | -0.00469 | | n_updates | 430 | | policy_gradient_loss | -0.00226 | | std | 0.909 | | value_loss | 0.0031 | ------------------------------------------ ------------------------------------------ | time/ | | | fps | 2129 | | iterations | 45 | | time_elapsed | 346 | | total_timesteps | 737280 | | train/ | | | approx_kl | 0.0063995067 | | clip_fraction | 0.0823 | | clip_range | 0.2 | | entropy_loss | -2.66 | | explained_variance | 0.951 | | learning_rate | 0.0003 | | loss | -0.0249 | | n_updates | 440 | | policy_gradient_loss | -0.00261 | | std | 0.922 | | value_loss | 0.00343 | ------------------------------------------ Eval num_timesteps=750000, episode_reward=-12.10 +/- 56.78 Episode length: 1850.50 +/- 449.09 ------------------------------------------ | eval/ | | | mean_ep_length | 1.85e+03 | | mean_reward | -12.1 | | time/ | | | total_timesteps | 750000 | | train/ | | | approx_kl | 0.0069549307 | | clip_fraction | 0.0847 | | clip_range | 0.2 | | entropy_loss | -2.68 | | explained_variance | 0.862 | | learning_rate | 0.0003 | | loss | -0.0192 | | n_updates | 450 | | policy_gradient_loss | -0.00165 | | std | 0.929 | | value_loss | 0.0032 | ------------------------------------------ [Diag @ 750,000 | n_sheep=2 | success=5%] COMPACT_CANT_DRIVE 9/20 NEVER_COMPACT 9/20 PARTIAL_1of2 1/20 SUCCESS 1/20 action_mag mean=0.261 p10=0.002 p90=0.983 (0=stopped, 1=full speed) min_flock_radius mean=3.93m best=0.00m (target <5m to compact) min_dog_to_com mean=0.79m best=0.07m (FLEE_DIST=7m) min_com_to_pen mean=13.43m best=1.62m reward/step (mean): progress=-0.0058 alignment=+0.0087 pen_bonus=+0.0008 step_cost=-0.0200 complete=+0.0025 ------------------------------- | time/ | | | fps | 2043 | | iterations | 46 | | time_elapsed | 368 | | total_timesteps | 753664 | ------------------------------- ----------------------------------------- | time/ | | | fps | 2062 | | iterations | 47 | | time_elapsed | 373 | | total_timesteps | 770048 | | train/ | | | approx_kl | 0.008165602 | | clip_fraction | 0.0997 | | clip_range | 0.2 | | entropy_loss | -2.69 | | explained_variance | 0.931 | | learning_rate | 0.0003 | | loss | -0.0461 | | n_updates | 460 | | policy_gradient_loss | -0.00412 | | std | 0.932 | | value_loss | 0.00308 | ----------------------------------------- ----------------------------------------- | time/ | | | fps | 2074 | | iterations | 48 | | time_elapsed | 379 | | total_timesteps | 786432 | | train/ | | | approx_kl | 0.006088208 | | clip_fraction | 0.0805 | | clip_range | 0.2 | | entropy_loss | -2.71 | | explained_variance | 0.917 | | learning_rate | 0.0003 | | loss | -0.034 | | n_updates | 470 | | policy_gradient_loss | -0.000257 | | std | 0.943 | | value_loss | 0.00533 | ----------------------------------------- Eval num_timesteps=800000, episode_reward=-32.78 +/- 23.33 Episode length: 2000.00 +/- 0.00 ------------------------------------------ | eval/ | | | mean_ep_length | 2e+03 | | mean_reward | -32.8 | | time/ | | | total_timesteps | 800000 | | train/ | | | approx_kl | 0.0069386996 | | clip_fraction | 0.0883 | | clip_range | 0.2 | | entropy_loss | -2.73 | | explained_variance | 0.954 | | learning_rate | 0.0003 | | loss | -0.0361 | | n_updates | 480 | | policy_gradient_loss | -0.00228 | | std | 0.948 | | value_loss | 0.00495 | ------------------------------------------ ------------------------------- | time/ | | | fps | 2028 | | iterations | 49 | | time_elapsed | 395 | | total_timesteps | 802816 | ------------------------------- ------------------------------------------ | time/ | | | fps | 2045 | | iterations | 50 | | time_elapsed | 400 | | total_timesteps | 819200 | | train/ | | | approx_kl | 0.0070893797 | | clip_fraction | 0.0687 | | clip_range | 0.2 | | entropy_loss | -2.74 | | explained_variance | 0.955 | | learning_rate | 0.0003 | | loss | -0.035 | | n_updates | 490 | | policy_gradient_loss | -0.00221 | | std | 0.954 | | value_loss | 0.00229 | ------------------------------------------ ------------------------------------------ | time/ | | | fps | 2060 | | iterations | 51 | | time_elapsed | 405 | | total_timesteps | 835584 | | train/ | | | approx_kl | 0.0068652867 | | clip_fraction | 0.0787 | | clip_range | 0.2 | | entropy_loss | -2.75 | | explained_variance | 0.863 | | learning_rate | 0.0003 | | loss | -0.0337 | | n_updates | 500 | | policy_gradient_loss | -0.00277 | | std | 0.959 | | value_loss | 0.00229 | ------------------------------------------ Eval num_timesteps=850000, episode_reward=-14.34 +/- 48.77 Episode length: 1998.40 +/- 6.97 ----------------------------------------- | eval/ | | | mean_ep_length | 2e+03 | | mean_reward | -14.3 | | time/ | | | total_timesteps | 850000 | | train/ | | | approx_kl | 0.007872021 | | clip_fraction | 0.0815 | | clip_range | 0.2 | | entropy_loss | -2.76 | | explained_variance | 0.852 | | learning_rate | 0.0003 | | loss | -0.0358 | | n_updates | 510 | | policy_gradient_loss | -0.00365 | | std | 0.966 | | value_loss | 0.00272 | ----------------------------------------- ------------------------------- | time/ | | | fps | 2018 | | iterations | 52 | | time_elapsed | 422 | | total_timesteps | 851968 | ------------------------------- ----------------------------------------- | time/ | | | fps | 2032 | | iterations | 53 | | time_elapsed | 427 | | total_timesteps | 868352 | | train/ | | | approx_kl | 0.007002457 | | clip_fraction | 0.0752 | | clip_range | 0.2 | | entropy_loss | -2.78 | | explained_variance | 0.879 | | learning_rate | 0.0003 | | loss | -0.0414 | | n_updates | 520 | | policy_gradient_loss | -0.00242 | | std | 0.977 | | value_loss | 0.00166 | ----------------------------------------- ----------------------------------------- | time/ | | | fps | 2047 | | iterations | 54 | | time_elapsed | 432 | | total_timesteps | 884736 | | train/ | | | approx_kl | 0.007822147 | | clip_fraction | 0.0813 | | clip_range | 0.2 | | entropy_loss | -2.8 | | explained_variance | 0.871 | | learning_rate | 0.0003 | | loss | -0.0376 | | n_updates | 530 | | policy_gradient_loss | -0.00362 | | std | 0.984 | | value_loss | 0.00212 | ----------------------------------------- Eval num_timesteps=900000, episode_reward=-20.41 +/- 60.01 Episode length: 1929.40 +/- 284.99 ---------------------------------------- | eval/ | | | mean_ep_length | 1.93e+03 | | mean_reward | -20.4 | | time/ | | | total_timesteps | 900000 | | train/ | | | approx_kl | 0.00738756 | | clip_fraction | 0.0793 | | clip_range | 0.2 | | entropy_loss | -2.81 | | explained_variance | 0.808 | | learning_rate | 0.0003 | | loss | -0.0355 | | n_updates | 540 | | policy_gradient_loss | -0.00195 | | std | 0.988 | | value_loss | 0.00721 | ---------------------------------------- [Diag @ 900,000 | n_sheep=2 | success=5%] COMPACT_CANT_DRIVE 11/20 NEVER_COMPACT 8/20 SUCCESS 1/20 action_mag mean=0.203 p10=0.007 p90=0.704 (0=stopped, 1=full speed) min_flock_radius mean=3.40m best=0.00m (target <5m to compact) min_dog_to_com mean=0.60m best=0.11m (FLEE_DIST=7m) min_com_to_pen mean=14.01m best=3.61m reward/step (mean): progress=-0.0040 alignment=+0.0071 pen_bonus=+0.0008 step_cost=-0.0200 complete=+0.0026 ------------------------------- | time/ | | | fps | 1977 | | iterations | 55 | | time_elapsed | 455 | | total_timesteps | 901120 | ------------------------------- ----------------------------------------- | time/ | | | fps | 1990 | | iterations | 56 | | time_elapsed | 460 | | total_timesteps | 917504 | | train/ | | | approx_kl | 0.007000256 | | clip_fraction | 0.0831 | | clip_range | 0.2 | | entropy_loss | -2.8 | | explained_variance | 0.889 | | learning_rate | 0.0003 | | loss | -0.0285 | | n_updates | 550 | | policy_gradient_loss | -0.00402 | | std | 0.984 | | value_loss | 0.00171 | ----------------------------------------- ----------------------------------------- | time/ | | | fps | 2005 | | iterations | 57 | | time_elapsed | 465 | | total_timesteps | 933888 | | train/ | | | approx_kl | 0.007749311 | | clip_fraction | 0.0755 | | clip_range | 0.2 | | entropy_loss | -2.83 | | explained_variance | 0.599 | | learning_rate | 0.0003 | | loss | -0.032 | | n_updates | 560 | | policy_gradient_loss | -0.00239 | | std | 1.01 | | value_loss | 0.00351 | ----------------------------------------- Eval num_timesteps=950000, episode_reward=-13.16 +/- 44.70 Episode length: 1949.30 +/- 221.00 ------------------------------------------ | eval/ | | | mean_ep_length | 1.95e+03 | | mean_reward | -13.2 | | time/ | | | total_timesteps | 950000 | | train/ | | | approx_kl | 0.0075328955 | | clip_fraction | 0.0829 | | clip_range | 0.2 | | entropy_loss | -2.85 | | explained_variance | 0.783 | | learning_rate | 0.0003 | | loss | -0.0306 | | n_updates | 570 | | policy_gradient_loss | -0.00352 | | std | 1.01 | | value_loss | 0.00319 | ------------------------------------------ ------------------------------- | time/ | | | fps | 1971 | | iterations | 58 | | time_elapsed | 482 | | total_timesteps | 950272 | ------------------------------- ------------------------------------------ | time/ | | | fps | 1981 | | iterations | 59 | | time_elapsed | 487 | | total_timesteps | 966656 | | train/ | | | approx_kl | 0.0072506005 | | clip_fraction | 0.0835 | | clip_range | 0.2 | | entropy_loss | -2.86 | | explained_variance | 0.929 | | learning_rate | 0.0003 | | loss | -0.0291 | | n_updates | 580 | | policy_gradient_loss | -0.00173 | | std | 1.01 | | value_loss | 0.00491 | ------------------------------------------ ------------------------------------------ | time/ | | | fps | 1991 | | iterations | 60 | | time_elapsed | 493 | | total_timesteps | 983040 | | train/ | | | approx_kl | 0.0068104668 | | clip_fraction | 0.0799 | | clip_range | 0.2 | | entropy_loss | -2.87 | | explained_variance | 0.813 | | learning_rate | 0.0003 | | loss | -0.0282 | | n_updates | 590 | | policy_gradient_loss | -0.00162 | | std | 1.02 | | value_loss | 0.00477 | ------------------------------------------ ----------------------------------------- | time/ | | | fps | 2005 | | iterations | 61 | | time_elapsed | 498 | | total_timesteps | 999424 | | train/ | | | approx_kl | 0.007103944 | | clip_fraction | 0.0774 | | clip_range | 0.2 | | entropy_loss | -2.88 | | explained_variance | 0.942 | | learning_rate | 0.0003 | | loss | -0.0322 | | n_updates | 600 | | policy_gradient_loss | -0.00143 | | std | 1.03 | | value_loss | 0.0033 | ----------------------------------------- Eval num_timesteps=1000000, episode_reward=-25.58 +/- 49.00 Episode length: 1999.50 +/- 2.18 ------------------------------------------ | eval/ | | | mean_ep_length | 2e+03 | | mean_reward | -25.6 | | time/ | | | total_timesteps | 1000000 | | train/ | | | approx_kl | 0.0075788023 | | clip_fraction | 0.088 | | clip_range | 0.2 | | entropy_loss | -2.9 | | explained_variance | 0.864 | | learning_rate | 0.0003 | | loss | -0.0352 | | n_updates | 610 | | policy_gradient_loss | -0.003 | | std | 1.04 | | value_loss | 0.00192 | ------------------------------------------ -------------------------------- | time/ | | | fps | 1971 | | iterations | 62 | | time_elapsed | 515 | | total_timesteps | 1015808 | -------------------------------- ----------------------------------------- | time/ | | | fps | 1983 | | iterations | 63 | | time_elapsed | 520 | | total_timesteps | 1032192 | | train/ | | | approx_kl | 0.009131588 | | clip_fraction | 0.0902 | | clip_range | 0.2 | | entropy_loss | -2.89 | | explained_variance | 0.941 | | learning_rate | 0.0003 | | loss | -0.0476 | | n_updates | 620 | | policy_gradient_loss | -0.00341 | | std | 1.03 | | value_loss | 0.00705 | ----------------------------------------- ---------------------------------------- | time/ | | | fps | 1995 | | iterations | 64 | | time_elapsed | 525 | | total_timesteps | 1048576 | | train/ | | | approx_kl | 0.00746674 | | clip_fraction | 0.0838 | | clip_range | 0.2 | | entropy_loss | -2.89 | | explained_variance | 0.958 | | learning_rate | 0.0003 | | loss | -0.022 | | n_updates | 630 | | policy_gradient_loss | -0.00392 | | std | 1.03 | | value_loss | 0.00592 | ---------------------------------------- Eval num_timesteps=1050000, episode_reward=-12.04 +/- 64.56 Episode length: 1889.90 +/- 333.38 ------------------------------------------ | eval/ | | | mean_ep_length | 1.89e+03 | | mean_reward | -12 | | time/ | | | total_timesteps | 1050000 | | train/ | | | approx_kl | 0.0058071706 | | clip_fraction | 0.0721 | | clip_range | 0.2 | | entropy_loss | -2.9 | | explained_variance | 0.932 | | learning_rate | 0.0003 | | loss | -0.0188 | | n_updates | 640 | | policy_gradient_loss | -0.00235 | | std | 1.03 | | value_loss | 0.00513 | ------------------------------------------ [Diag @ 1,050,000 | n_sheep=2 | success=5%] COMPACT_CANT_DRIVE 10/20 NEVER_COMPACT 9/20 SUCCESS 1/20 action_mag mean=0.190 p10=0.001 p90=0.686 (0=stopped, 1=full speed) min_flock_radius mean=4.60m best=0.00m (target <5m to compact) min_dog_to_com mean=0.54m best=0.21m (FLEE_DIST=7m) min_com_to_pen mean=13.05m best=3.62m reward/step (mean): progress=-0.0023 alignment=+0.0072 pen_bonus=+0.0005 step_cost=-0.0200 complete=+0.0025 -------------------------------- | time/ | | | fps | 1931 | | iterations | 65 | | time_elapsed | 551 | | total_timesteps | 1064960 | -------------------------------- ----------------------------------------- | time/ | | | fps | 1944 | | iterations | 66 | | time_elapsed | 556 | | total_timesteps | 1081344 | | train/ | | | approx_kl | 0.006802067 | | clip_fraction | 0.0701 | | clip_range | 0.2 | | entropy_loss | -2.92 | | explained_variance | 0.937 | | learning_rate | 0.0003 | | loss | -0.0304 | | n_updates | 650 | | policy_gradient_loss | -0.0019 | | std | 1.04 | | value_loss | 0.00206 | ----------------------------------------- ----------------------------------------- | time/ | | | fps | 1956 | | iterations | 67 | | time_elapsed | 561 | | total_timesteps | 1097728 | | train/ | | | approx_kl | 0.007102525 | | clip_fraction | 0.074 | | clip_range | 0.2 | | entropy_loss | -2.92 | | explained_variance | 0.953 | | learning_rate | 0.0003 | | loss | -0.00869 | | n_updates | 660 | | policy_gradient_loss | -0.00208 | | std | 1.04 | | value_loss | 0.00579 | ----------------------------------------- Eval num_timesteps=1100000, episode_reward=-29.51 +/- 23.80 Episode length: 2000.00 +/- 0.00 ----------------------------------------- | eval/ | | | mean_ep_length | 2e+03 | | mean_reward | -29.5 | | time/ | | | total_timesteps | 1100000 | | train/ | | | approx_kl | 0.006372301 | | clip_fraction | 0.0669 | | clip_range | 0.2 | | entropy_loss | -2.94 | | explained_variance | 0.829 | | learning_rate | 0.0003 | | loss | -0.0349 | | n_updates | 670 | | policy_gradient_loss | -0.00135 | | std | 1.06 | | value_loss | 0.00208 | ----------------------------------------- -------------------------------- | time/ | | | fps | 1932 | | iterations | 68 | | time_elapsed | 576 | | total_timesteps | 1114112 | -------------------------------- ----------------------------------------- | time/ | | | fps | 1942 | | iterations | 69 | | time_elapsed | 581 | | total_timesteps | 1130496 | | train/ | | | approx_kl | 0.007083354 | | clip_fraction | 0.0839 | | clip_range | 0.2 | | entropy_loss | -2.95 | | explained_variance | 0.845 | | learning_rate | 0.0003 | | loss | -0.0464 | | n_updates | 680 | | policy_gradient_loss | -0.00298 | | std | 1.06 | | value_loss | 0.00747 | ----------------------------------------- ----------------------------------------- | time/ | | | fps | 1954 | | iterations | 70 | | time_elapsed | 586 | | total_timesteps | 1146880 | | train/ | | | approx_kl | 0.007034454 | | clip_fraction | 0.0875 | | clip_range | 0.2 | | entropy_loss | -2.96 | | explained_variance | 0.892 | | learning_rate | 0.0003 | | loss | -0.0382 | | n_updates | 690 | | policy_gradient_loss | -0.00359 | | std | 1.06 | | value_loss | 0.00208 | ----------------------------------------- Eval num_timesteps=1150000, episode_reward=-20.98 +/- 49.18 Episode length: 1959.70 +/- 175.66 ----------------------------------------- | eval/ | | | mean_ep_length | 1.96e+03 | | mean_reward | -21 | | time/ | | | total_timesteps | 1150000 | | train/ | | | approx_kl | 0.006192833 | | clip_fraction | 0.0626 | | clip_range | 0.2 | | entropy_loss | -2.94 | | explained_variance | 0.951 | | learning_rate | 0.0003 | | loss | -0.0224 | | n_updates | 700 | | policy_gradient_loss | -0.00299 | | std | 1.05 | | value_loss | 0.00883 | ----------------------------------------- -------------------------------- | time/ | | | fps | 1926 | | iterations | 71 | | time_elapsed | 603 | | total_timesteps | 1163264 | -------------------------------- ----------------------------------------- | time/ | | | fps | 1937 | | iterations | 72 | | time_elapsed | 608 | | total_timesteps | 1179648 | | train/ | | | approx_kl | 0.008185772 | | clip_fraction | 0.0969 | | clip_range | 0.2 | | entropy_loss | -2.96 | | explained_variance | 0.944 | | learning_rate | 0.0003 | | loss | -0.0278 | | n_updates | 710 | | policy_gradient_loss | -0.00316 | | std | 1.07 | | value_loss | 0.00421 | ----------------------------------------- ------------------------------------------ | time/ | | | fps | 1947 | | iterations | 73 | | time_elapsed | 614 | | total_timesteps | 1196032 | | train/ | | | approx_kl | 0.0063469247 | | clip_fraction | 0.065 | | clip_range | 0.2 | | entropy_loss | -2.96 | | explained_variance | 0.912 | | learning_rate | 0.0003 | | loss | -0.0239 | | n_updates | 720 | | policy_gradient_loss | -0.00224 | | std | 1.06 | | value_loss | 0.0054 | ------------------------------------------ Eval num_timesteps=1200000, episode_reward=-29.34 +/- 18.71 Episode length: 2000.00 +/- 0.00 ---------------------------------------- | eval/ | | | mean_ep_length | 2e+03 | | mean_reward | -29.3 | | time/ | | | total_timesteps | 1200000 | | train/ | | | approx_kl | 0.00778389 | | clip_fraction | 0.0734 | | clip_range | 0.2 | | entropy_loss | -2.95 | | explained_variance | 0.961 | | learning_rate | 0.0003 | | loss | -0.0435 | | n_updates | 730 | | policy_gradient_loss | -0.00184 | | std | 1.06 | | value_loss | 0.0048 | ---------------------------------------- [Diag @ 1,200,000 | n_sheep=2 | success=10%] NEVER_COMPACT 9/20 COMPACT_CANT_DRIVE 9/20 SUCCESS 2/20 action_mag mean=0.198 p10=0.002 p90=0.744 (0=stopped, 1=full speed) min_flock_radius mean=3.94m best=0.00m (target <5m to compact) min_dog_to_com mean=0.50m best=0.14m (FLEE_DIST=7m) min_com_to_pen mean=11.36m best=3.58m reward/step (mean): progress=-0.0002 alignment=+0.0073 pen_bonus=+0.0013 step_cost=-0.0200 complete=+0.0053 [Curriculum] leaving stage n_sheep=2 after 600,000 steps | training success rate (last 100 eps) = 5% [Curriculum] → 3 sheep at step 1,200,000 -------------------------------- | time/ | | | fps | 1898 | | iterations | 74 | | time_elapsed | 638 | | total_timesteps | 1212416 | -------------------------------- ---------------------------------------- | time/ | | | fps | 1909 | | iterations | 75 | | time_elapsed | 643 | | total_timesteps | 1228800 | | train/ | | | approx_kl | 0.00918101 | | clip_fraction | 0.106 | | clip_range | 0.2 | | entropy_loss | -2.95 | | explained_variance | 0.919 | | learning_rate | 0.0003 | | loss | -0.0112 | | n_updates | 740 | | policy_gradient_loss | -0.00123 | | std | 1.06 | | value_loss | 0.0427 | ---------------------------------------- ----------------------------------------- | time/ | | | fps | 1917 | | iterations | 76 | | time_elapsed | 649 | | total_timesteps | 1245184 | | train/ | | | approx_kl | 0.010076641 | | clip_fraction | 0.137 | | clip_range | 0.2 | | entropy_loss | -2.94 | | explained_variance | 0.919 | | learning_rate | 0.0003 | | loss | -0.0229 | | n_updates | 750 | | policy_gradient_loss | -0.000617 | | std | 1.05 | | value_loss | 0.0222 | ----------------------------------------- Eval num_timesteps=1250000, episode_reward=-38.73 +/- 33.85 Episode length: 2000.00 +/- 0.00 --------------------------------------- | eval/ | | | mean_ep_length | 2e+03 | | mean_reward | -38.7 | | time/ | | | total_timesteps | 1250000 | | train/ | | | approx_kl | 0.0084493 | | clip_fraction | 0.109 | | clip_range | 0.2 | | entropy_loss | -2.96 | | explained_variance | 0.96 | | learning_rate | 0.0003 | | loss | -0.0259 | | n_updates | 760 | | policy_gradient_loss | -0.00168 | | std | 1.06 | | value_loss | 0.0024 | --------------------------------------- -------------------------------- | time/ | | | fps | 1890 | | iterations | 77 | | time_elapsed | 667 | | total_timesteps | 1261568 | -------------------------------- ----------------------------------------- | time/ | | | fps | 1899 | | iterations | 78 | | time_elapsed | 672 | | total_timesteps | 1277952 | | train/ | | | approx_kl | 0.008724872 | | clip_fraction | 0.109 | | clip_range | 0.2 | | entropy_loss | -2.98 | | explained_variance | 0.931 | | learning_rate | 0.0003 | | loss | -0.0293 | | n_updates | 770 | | policy_gradient_loss | -0.00204 | | std | 1.08 | | value_loss | 0.0067 | ----------------------------------------- ----------------------------------------- | time/ | | | fps | 1906 | | iterations | 79 | | time_elapsed | 678 | | total_timesteps | 1294336 | | train/ | | | approx_kl | 0.008191848 | | clip_fraction | 0.096 | | clip_range | 0.2 | | entropy_loss | -2.99 | | explained_variance | 0.963 | | learning_rate | 0.0003 | | loss | -0.0247 | | n_updates | 780 | | policy_gradient_loss | -0.002 | | std | 1.08 | | value_loss | 0.00632 | ----------------------------------------- Eval num_timesteps=1300000, episode_reward=-26.68 +/- 27.12 Episode length: 2000.00 +/- 0.00 ----------------------------------------- | eval/ | | | mean_ep_length | 2e+03 | | mean_reward | -26.7 | | time/ | | | total_timesteps | 1300000 | | train/ | | | approx_kl | 0.006018152 | | clip_fraction | 0.0869 | | clip_range | 0.2 | | entropy_loss | -3 | | explained_variance | 0.96 | | learning_rate | 0.0003 | | loss | -0.0311 | | n_updates | 790 | | policy_gradient_loss | -0.00129 | | std | 1.09 | | value_loss | 0.00189 | ----------------------------------------- -------------------------------- | time/ | | | fps | 1881 | | iterations | 80 | | time_elapsed | 696 | | total_timesteps | 1310720 | -------------------------------- ------------------------------------------ | time/ | | | fps | 1892 | | iterations | 81 | | time_elapsed | 701 | | total_timesteps | 1327104 | | train/ | | | approx_kl | 0.0077671953 | | clip_fraction | 0.082 | | clip_range | 0.2 | | entropy_loss | -3.01 | | explained_variance | 0.972 | | learning_rate | 0.0003 | | loss | -0.0308 | | n_updates | 800 | | policy_gradient_loss | -0.00219 | | std | 1.09 | | value_loss | 0.00177 | ------------------------------------------ ----------------------------------------- | time/ | | | fps | 1902 | | iterations | 82 | | time_elapsed | 706 | | total_timesteps | 1343488 | | train/ | | | approx_kl | 0.008806022 | | clip_fraction | 0.0947 | | clip_range | 0.2 | | entropy_loss | -3.02 | | explained_variance | 0.962 | | learning_rate | 0.0003 | | loss | -0.0426 | | n_updates | 810 | | policy_gradient_loss | -0.00231 | | std | 1.1 | | value_loss | 0.00235 | ----------------------------------------- Eval num_timesteps=1350000, episode_reward=-24.30 +/- 32.03 Episode length: 2000.00 +/- 0.00 ----------------------------------------- | eval/ | | | mean_ep_length | 2e+03 | | mean_reward | -24.3 | | time/ | | | total_timesteps | 1350000 | | train/ | | | approx_kl | 0.007263833 | | clip_fraction | 0.0797 | | clip_range | 0.2 | | entropy_loss | -3.03 | | explained_variance | 0.957 | | learning_rate | 0.0003 | | loss | -0.0338 | | n_updates | 820 | | policy_gradient_loss | -0.00251 | | std | 1.11 | | value_loss | 0.00397 | ----------------------------------------- [Diag @ 1,350,000 | n_sheep=3 | success=0%] NEVER_COMPACT 16/20 COMPACT_CANT_DRIVE 4/20 action_mag mean=0.058 p10=0.004 p90=0.054 (0=stopped, 1=full speed) min_flock_radius mean=6.77m best=1.04m (target <5m to compact) min_dog_to_com mean=0.58m best=0.28m (FLEE_DIST=7m) min_com_to_pen mean=12.71m best=4.27m reward/step (mean): progress=-0.0038 alignment=+0.0015 pen_bonus=+0.0005 step_cost=-0.0200 complete=+0.0000 -------------------------------- | time/ | | | fps | 1859 | | iterations | 83 | | time_elapsed | 731 | | total_timesteps | 1359872 | -------------------------------- ----------------------------------------- | time/ | | | fps | 1870 | | iterations | 84 | | time_elapsed | 735 | | total_timesteps | 1376256 | | train/ | | | approx_kl | 0.007816839 | | clip_fraction | 0.0812 | | clip_range | 0.2 | | entropy_loss | -3.05 | | explained_variance | 0.946 | | learning_rate | 0.0003 | | loss | -0.0285 | | n_updates | 830 | | policy_gradient_loss | -0.00277 | | std | 1.11 | | value_loss | 0.0018 | ----------------------------------------- ------------------------------------------ | time/ | | | fps | 1880 | | iterations | 85 | | time_elapsed | 740 | | total_timesteps | 1392640 | | train/ | | | approx_kl | 0.0064534983 | | clip_fraction | 0.0774 | | clip_range | 0.2 | | entropy_loss | -3.06 | | explained_variance | 0.958 | | learning_rate | 0.0003 | | loss | -0.0305 | | n_updates | 840 | | policy_gradient_loss | -0.00158 | | std | 1.12 | | value_loss | 0.00988 | ------------------------------------------ Eval num_timesteps=1400000, episode_reward=-39.10 +/- 41.08 Episode length: 2000.00 +/- 0.00 ------------------------------------------ | eval/ | | | mean_ep_length | 2e+03 | | mean_reward | -39.1 | | time/ | | | total_timesteps | 1400000 | | train/ | | | approx_kl | 0.0069560152 | | clip_fraction | 0.0835 | | clip_range | 0.2 | | entropy_loss | -3.07 | | explained_variance | 0.96 | | learning_rate | 0.0003 | | loss | -0.0302 | | n_updates | 850 | | policy_gradient_loss | -0.00283 | | std | 1.12 | | value_loss | 0.00307 | ------------------------------------------ -------------------------------- | time/ | | | fps | 1857 | | iterations | 86 | | time_elapsed | 758 | | total_timesteps | 1409024 | -------------------------------- ----------------------------------------- | time/ | | | fps | 1864 | | iterations | 87 | | time_elapsed | 764 | | total_timesteps | 1425408 | | train/ | | | approx_kl | 0.007682803 | | clip_fraction | 0.0931 | | clip_range | 0.2 | | entropy_loss | -3.09 | | explained_variance | 0.902 | | learning_rate | 0.0003 | | loss | -0.0322 | | n_updates | 860 | | policy_gradient_loss | -0.00224 | | std | 1.14 | | value_loss | 0.013 | ----------------------------------------- ------------------------------------------ | time/ | | | fps | 1869 | | iterations | 88 | | time_elapsed | 771 | | total_timesteps | 1441792 | | train/ | | | approx_kl | 0.0063949013 | | clip_fraction | 0.0786 | | clip_range | 0.2 | | entropy_loss | -3.1 | | explained_variance | 0.953 | | learning_rate | 0.0003 | | loss | -0.0401 | | n_updates | 870 | | policy_gradient_loss | -0.00134 | | std | 1.14 | | value_loss | 0.00193 | ------------------------------------------ Eval num_timesteps=1450000, episode_reward=-28.59 +/- 25.61 Episode length: 2000.00 +/- 0.00 ----------------------------------------- | eval/ | | | mean_ep_length | 2e+03 | | mean_reward | -28.6 | | time/ | | | total_timesteps | 1450000 | | train/ | | | approx_kl | 0.007503539 | | clip_fraction | 0.0774 | | clip_range | 0.2 | | entropy_loss | -3.13 | | explained_variance | 0.951 | | learning_rate | 0.0003 | | loss | -0.0378 | | n_updates | 880 | | policy_gradient_loss | -0.00309 | | std | 1.16 | | value_loss | 0.00551 | ----------------------------------------- -------------------------------- | time/ | | | fps | 1845 | | iterations | 89 | | time_elapsed | 789 | | total_timesteps | 1458176 | -------------------------------- ------------------------------------------ | time/ | | | fps | 1852 | | iterations | 90 | | time_elapsed | 796 | | total_timesteps | 1474560 | | train/ | | | approx_kl | 0.0075057503 | | clip_fraction | 0.0793 | | clip_range | 0.2 | | entropy_loss | -3.15 | | explained_variance | 0.955 | | learning_rate | 0.0003 | | loss | -0.0439 | | n_updates | 890 | | policy_gradient_loss | -0.00264 | | std | 1.17 | | value_loss | 0.00265 | ------------------------------------------ ------------------------------------------ | time/ | | | fps | 1857 | | iterations | 91 | | time_elapsed | 802 | | total_timesteps | 1490944 | | train/ | | | approx_kl | 0.0068523246 | | clip_fraction | 0.0755 | | clip_range | 0.2 | | entropy_loss | -3.15 | | explained_variance | 0.935 | | learning_rate | 0.0003 | | loss | -0.0282 | | n_updates | 900 | | policy_gradient_loss | -0.00292 | | std | 1.17 | | value_loss | 0.00268 | ------------------------------------------ Eval num_timesteps=1500000, episode_reward=-40.66 +/- 25.29 Episode length: 2000.00 +/- 0.00 ----------------------------------------- | eval/ | | | mean_ep_length | 2e+03 | | mean_reward | -40.7 | | time/ | | | total_timesteps | 1500000 | | train/ | | | approx_kl | 0.007249858 | | clip_fraction | 0.0857 | | clip_range | 0.2 | | entropy_loss | -3.15 | | explained_variance | 0.952 | | learning_rate | 0.0003 | | loss | -0.0366 | | n_updates | 910 | | policy_gradient_loss | -0.00319 | | std | 1.17 | | value_loss | 0.00564 | ----------------------------------------- [Diag @ 1,500,000 | n_sheep=3 | success=0%] NEVER_COMPACT 14/20 COMPACT_CANT_DRIVE 6/20 action_mag mean=0.050 p10=0.005 p90=0.049 (0=stopped, 1=full speed) min_flock_radius mean=6.53m best=0.98m (target <5m to compact) min_dog_to_com mean=0.46m best=0.06m (FLEE_DIST=7m) min_com_to_pen mean=12.38m best=5.44m reward/step (mean): progress=+0.0039 alignment=+0.0011 pen_bonus=+0.0005 step_cost=-0.0200 complete=+0.0000 -------------------------------- | time/ | | | fps | 1819 | | iterations | 92 | | time_elapsed | 828 | | total_timesteps | 1507328 | -------------------------------- ----------------------------------------- | time/ | | | fps | 1828 | | iterations | 93 | | time_elapsed | 833 | | total_timesteps | 1523712 | | train/ | | | approx_kl | 0.007471386 | | clip_fraction | 0.0834 | | clip_range | 0.2 | | entropy_loss | -3.16 | | explained_variance | 0.929 | | learning_rate | 0.0003 | | loss | -0.0275 | | n_updates | 920 | | policy_gradient_loss | -0.00192 | | std | 1.17 | | value_loss | 0.00791 | ----------------------------------------- ----------------------------------------- | time/ | | | fps | 1835 | | iterations | 94 | | time_elapsed | 838 | | total_timesteps | 1540096 | | train/ | | | approx_kl | 0.007296456 | | clip_fraction | 0.0765 | | clip_range | 0.2 | | entropy_loss | -3.17 | | explained_variance | 0.95 | | learning_rate | 0.0003 | | loss | -0.0484 | | n_updates | 930 | | policy_gradient_loss | -0.00366 | | std | 1.18 | | value_loss | 0.00788 | ----------------------------------------- Eval num_timesteps=1550000, episode_reward=-34.66 +/- 25.47 Episode length: 2000.00 +/- 0.00 ----------------------------------------- | eval/ | | | mean_ep_length | 2e+03 | | mean_reward | -34.7 | | time/ | | | total_timesteps | 1550000 | | train/ | | | approx_kl | 0.007654687 | | clip_fraction | 0.095 | | clip_range | 0.2 | | entropy_loss | -3.18 | | explained_variance | 0.92 | | learning_rate | 0.0003 | | loss | -0.0386 | | n_updates | 940 | | policy_gradient_loss | -0.00316 | | std | 1.19 | | value_loss | 0.00363 | ----------------------------------------- -------------------------------- | time/ | | | fps | 1817 | | iterations | 95 | | time_elapsed | 856 | | total_timesteps | 1556480 | -------------------------------- ----------------------------------------- | time/ | | | fps | 1823 | | iterations | 96 | | time_elapsed | 862 | | total_timesteps | 1572864 | | train/ | | | approx_kl | 0.007030643 | | clip_fraction | 0.0881 | | clip_range | 0.2 | | entropy_loss | -3.18 | | explained_variance | 0.944 | | learning_rate | 0.0003 | | loss | -0.0346 | | n_updates | 950 | | policy_gradient_loss | -0.00321 | | std | 1.19 | | value_loss | 0.00208 | ----------------------------------------- ------------------------------------------ | time/ | | | fps | 1828 | | iterations | 97 | | time_elapsed | 869 | | total_timesteps | 1589248 | | train/ | | | approx_kl | 0.0071562277 | | clip_fraction | 0.0834 | | clip_range | 0.2 | | entropy_loss | -3.19 | | explained_variance | 0.955 | | learning_rate | 0.0003 | | loss | -0.0196 | | n_updates | 960 | | policy_gradient_loss | -0.00259 | | std | 1.2 | | value_loss | 0.00773 | ------------------------------------------ Eval num_timesteps=1600000, episode_reward=-33.49 +/- 36.88 Episode length: 2000.00 +/- 0.00 ------------------------------------------ | eval/ | | | mean_ep_length | 2e+03 | | mean_reward | -33.5 | | time/ | | | total_timesteps | 1600000 | | train/ | | | approx_kl | 0.0069667175 | | clip_fraction | 0.0741 | | clip_range | 0.2 | | entropy_loss | -3.2 | | explained_variance | 0.94 | | learning_rate | 0.0003 | | loss | -0.0313 | | n_updates | 970 | | policy_gradient_loss | -0.00399 | | std | 1.2 | | value_loss | 0.00419 | ------------------------------------------ -------------------------------- | time/ | | | fps | 1810 | | iterations | 98 | | time_elapsed | 886 | | total_timesteps | 1605632 | -------------------------------- ------------------------------------------ | time/ | | | fps | 1819 | | iterations | 99 | | time_elapsed | 891 | | total_timesteps | 1622016 | | train/ | | | approx_kl | 0.0061995042 | | clip_fraction | 0.0767 | | clip_range | 0.2 | | entropy_loss | -3.21 | | explained_variance | 0.968 | | learning_rate | 0.0003 | | loss | -0.036 | | n_updates | 980 | | policy_gradient_loss | -0.00289 | | std | 1.2 | | value_loss | 0.00241 | ------------------------------------------ ----------------------------------------- | time/ | | | fps | 1826 | | iterations | 100 | | time_elapsed | 896 | | total_timesteps | 1638400 | | train/ | | | approx_kl | 0.006502889 | | clip_fraction | 0.0714 | | clip_range | 0.2 | | entropy_loss | -3.22 | | explained_variance | 0.976 | | learning_rate | 0.0003 | | loss | -0.0445 | | n_updates | 990 | | policy_gradient_loss | -0.00314 | | std | 1.21 | | value_loss | 0.00218 | ----------------------------------------- Eval num_timesteps=1650000, episode_reward=-38.00 +/- 30.02 Episode length: 2000.00 +/- 0.00 ----------------------------------------- | eval/ | | | mean_ep_length | 2e+03 | | mean_reward | -38 | | time/ | | | total_timesteps | 1650000 | | train/ | | | approx_kl | 0.006163503 | | clip_fraction | 0.0739 | | clip_range | 0.2 | | entropy_loss | -3.22 | | explained_variance | 0.955 | | learning_rate | 0.0003 | | loss | -0.0391 | | n_updates | 1000 | | policy_gradient_loss | -0.00257 | | std | 1.22 | | value_loss | 0.0027 | ----------------------------------------- [Diag @ 1,650,000 | n_sheep=3 | success=0%] NEVER_COMPACT 16/20 COMPACT_CANT_DRIVE 4/20 action_mag mean=0.054 p10=0.002 p90=0.051 (0=stopped, 1=full speed) min_flock_radius mean=6.63m best=3.72m (target <5m to compact) min_dog_to_com mean=0.60m best=0.09m (FLEE_DIST=7m) min_com_to_pen mean=13.17m best=5.44m reward/step (mean): progress=+0.0032 alignment=+0.0015 pen_bonus=+0.0005 step_cost=-0.0200 complete=+0.0000 -------------------------------- | time/ | | | fps | 1793 | | iterations | 101 | | time_elapsed | 922 | | total_timesteps | 1654784 | -------------------------------- ---------------------------------------- | time/ | | | fps | 1800 | | iterations | 102 | | time_elapsed | 927 | | total_timesteps | 1671168 | | train/ | | | approx_kl | 0.00634938 | | clip_fraction | 0.073 | | clip_range | 0.2 | | entropy_loss | -3.23 | | explained_variance | 0.97 | | learning_rate | 0.0003 | | loss | -0.0462 | | n_updates | 1010 | | policy_gradient_loss | -0.00394 | | std | 1.22 | | value_loss | 0.00334 | ---------------------------------------- ------------------------------------------ | time/ | | | fps | 1807 | | iterations | 103 | | time_elapsed | 933 | | total_timesteps | 1687552 | | train/ | | | approx_kl | 0.0072235917 | | clip_fraction | 0.0774 | | clip_range | 0.2 | | entropy_loss | -3.23 | | explained_variance | 0.957 | | learning_rate | 0.0003 | | loss | -0.0284 | | n_updates | 1020 | | policy_gradient_loss | -0.00292 | | std | 1.22 | | value_loss | 0.00807 | ------------------------------------------ Eval num_timesteps=1700000, episode_reward=-32.26 +/- 31.96 Episode length: 2000.00 +/- 0.00 ------------------------------------------ | eval/ | | | mean_ep_length | 2e+03 | | mean_reward | -32.3 | | time/ | | | total_timesteps | 1700000 | | train/ | | | approx_kl | 0.0060304543 | | clip_fraction | 0.0721 | | clip_range | 0.2 | | entropy_loss | -3.23 | | explained_variance | 0.929 | | learning_rate | 0.0003 | | loss | -0.0427 | | n_updates | 1030 | | policy_gradient_loss | -0.00306 | | std | 1.21 | | value_loss | 0.00208 | ------------------------------------------ -------------------------------- | time/ | | | fps | 1789 | | iterations | 104 | | time_elapsed | 952 | | total_timesteps | 1703936 | -------------------------------- ----------------------------------------- | time/ | | | fps | 1795 | | iterations | 105 | | time_elapsed | 958 | | total_timesteps | 1720320 | | train/ | | | approx_kl | 0.006440907 | | clip_fraction | 0.0642 | | clip_range | 0.2 | | entropy_loss | -3.22 | | explained_variance | 0.947 | | learning_rate | 0.0003 | | loss | -0.0317 | | n_updates | 1040 | | policy_gradient_loss | -0.00158 | | std | 1.21 | | value_loss | 0.00165 | ----------------------------------------- ----------------------------------------- | time/ | | | fps | 1801 | | iterations | 106 | | time_elapsed | 963 | | total_timesteps | 1736704 | | train/ | | | approx_kl | 0.006897255 | | clip_fraction | 0.0738 | | clip_range | 0.2 | | entropy_loss | -3.2 | | explained_variance | 0.939 | | learning_rate | 0.0003 | | loss | -0.0408 | | n_updates | 1050 | | policy_gradient_loss | -0.00349 | | std | 1.19 | | value_loss | 0.00814 | ----------------------------------------- Eval num_timesteps=1750000, episode_reward=-40.58 +/- 28.91 Episode length: 2000.00 +/- 0.00 ------------------------------------------ | eval/ | | | mean_ep_length | 2e+03 | | mean_reward | -40.6 | | time/ | | | total_timesteps | 1750000 | | train/ | | | approx_kl | 0.0070952754 | | clip_fraction | 0.0742 | | clip_range | 0.2 | | entropy_loss | -3.19 | | explained_variance | 0.957 | | learning_rate | 0.0003 | | loss | -0.0308 | | n_updates | 1060 | | policy_gradient_loss | -0.0037 | | std | 1.19 | | value_loss | 0.0191 | ------------------------------------------ -------------------------------- | time/ | | | fps | 1784 | | iterations | 107 | | time_elapsed | 982 | | total_timesteps | 1753088 | -------------------------------- ----------------------------------------- | time/ | | | fps | 1791 | | iterations | 108 | | time_elapsed | 987 | | total_timesteps | 1769472 | | train/ | | | approx_kl | 0.006444447 | | clip_fraction | 0.0736 | | clip_range | 0.2 | | entropy_loss | -3.2 | | explained_variance | 0.968 | | learning_rate | 0.0003 | | loss | -0.0362 | | n_updates | 1070 | | policy_gradient_loss | -0.00409 | | std | 1.2 | | value_loss | 0.00395 | ----------------------------------------- ----------------------------------------- | time/ | | | fps | 1797 | | iterations | 109 | | time_elapsed | 993 | | total_timesteps | 1785856 | | train/ | | | approx_kl | 0.007391736 | | clip_fraction | 0.0758 | | clip_range | 0.2 | | entropy_loss | -3.22 | | explained_variance | 0.96 | | learning_rate | 0.0003 | | loss | -0.0341 | | n_updates | 1080 | | policy_gradient_loss | -0.00272 | | std | 1.21 | | value_loss | 0.00221 | ----------------------------------------- Eval num_timesteps=1800000, episode_reward=-29.06 +/- 30.98 Episode length: 2000.00 +/- 0.00 ----------------------------------------- | eval/ | | | mean_ep_length | 2e+03 | | mean_reward | -29.1 | | time/ | | | total_timesteps | 1800000 | | train/ | | | approx_kl | 0.006899439 | | clip_fraction | 0.0695 | | clip_range | 0.2 | | entropy_loss | -3.25 | | explained_variance | 0.965 | | learning_rate | 0.0003 | | loss | -0.0317 | | n_updates | 1090 | | policy_gradient_loss | -0.00226 | | std | 1.23 | | value_loss | 0.00615 | ----------------------------------------- [Diag @ 1,800,000 | n_sheep=3 | success=0%] NEVER_COMPACT 11/20 COMPACT_CANT_DRIVE 9/20 action_mag mean=0.054 p10=0.003 p90=0.057 (0=stopped, 1=full speed) min_flock_radius mean=6.01m best=1.13m (target <5m to compact) min_dog_to_com mean=0.51m best=0.11m (FLEE_DIST=7m) min_com_to_pen mean=12.52m best=3.21m reward/step (mean): progress=+0.0050 alignment=+0.0017 pen_bonus=+0.0008 step_cost=-0.0200 complete=+0.0000 [Curriculum] leaving stage n_sheep=3 after 600,000 steps | training success rate (last 100 eps) = 0% [Curriculum] → 4 sheep at step 1,800,000 -------------------------------- | time/ | | | fps | 1769 | | iterations | 110 | | time_elapsed | 1018 | | total_timesteps | 1802240 | -------------------------------- ----------------------------------------- | time/ | | | fps | 1776 | | iterations | 111 | | time_elapsed | 1023 | | total_timesteps | 1818624 | | train/ | | | approx_kl | 0.006710761 | | clip_fraction | 0.0761 | | clip_range | 0.2 | | entropy_loss | -3.25 | | explained_variance | 0.867 | | learning_rate | 0.0003 | | loss | -0.031 | | n_updates | 1100 | | policy_gradient_loss | -0.00311 | | std | 1.23 | | value_loss | 0.0186 | ----------------------------------------- ----------------------------------------- | time/ | | | fps | 1783 | | iterations | 112 | | time_elapsed | 1028 | | total_timesteps | 1835008 | | train/ | | | approx_kl | 0.006202608 | | clip_fraction | 0.0682 | | clip_range | 0.2 | | entropy_loss | -3.25 | | explained_variance | 0.954 | | learning_rate | 0.0003 | | loss | -0.0245 | | n_updates | 1110 | | policy_gradient_loss | -0.00429 | | std | 1.23 | | value_loss | 0.00641 | ----------------------------------------- Eval num_timesteps=1850000, episode_reward=-35.87 +/- 42.36 Episode length: 2000.00 +/- 0.00 ----------------------------------------- | eval/ | | | mean_ep_length | 2e+03 | | mean_reward | -35.9 | | time/ | | | total_timesteps | 1850000 | | train/ | | | approx_kl | 0.008398036 | | clip_fraction | 0.086 | | clip_range | 0.2 | | entropy_loss | -3.28 | | explained_variance | 0.938 | | learning_rate | 0.0003 | | loss | -0.0514 | | n_updates | 1120 | | policy_gradient_loss | -0.00497 | | std | 1.25 | | value_loss | 0.00614 | ----------------------------------------- -------------------------------- | time/ | | | fps | 1768 | | iterations | 113 | | time_elapsed | 1046 | | total_timesteps | 1851392 | -------------------------------- ----------------------------------------- | time/ | | | fps | 1775 | | iterations | 114 | | time_elapsed | 1052 | | total_timesteps | 1867776 | | train/ | | | approx_kl | 0.007641702 | | clip_fraction | 0.0742 | | clip_range | 0.2 | | entropy_loss | -3.31 | | explained_variance | 0.935 | | learning_rate | 0.0003 | | loss | -0.046 | | n_updates | 1130 | | policy_gradient_loss | -0.00349 | | std | 1.28 | | value_loss | 0.0228 | ----------------------------------------- ------------------------------------------ | time/ | | | fps | 1781 | | iterations | 115 | | time_elapsed | 1057 | | total_timesteps | 1884160 | | train/ | | | approx_kl | 0.0073437546 | | clip_fraction | 0.0747 | | clip_range | 0.2 | | entropy_loss | -3.34 | | explained_variance | 0.928 | | learning_rate | 0.0003 | | loss | -0.0498 | | n_updates | 1140 | | policy_gradient_loss | -0.00496 | | std | 1.29 | | value_loss | 0.00764 | ------------------------------------------ Eval num_timesteps=1900000, episode_reward=-41.88 +/- 27.01 Episode length: 2000.00 +/- 0.00 ----------------------------------------- | eval/ | | | mean_ep_length | 2e+03 | | mean_reward | -41.9 | | time/ | | | total_timesteps | 1900000 | | train/ | | | approx_kl | 0.006885264 | | clip_fraction | 0.0728 | | clip_range | 0.2 | | entropy_loss | -3.36 | | explained_variance | 0.934 | | learning_rate | 0.0003 | | loss | -0.0503 | | n_updates | 1150 | | policy_gradient_loss | -0.00384 | | std | 1.3 | | value_loss | 0.00423 | ----------------------------------------- -------------------------------- | time/ | | | fps | 1767 | | iterations | 116 | | time_elapsed | 1075 | | total_timesteps | 1900544 | -------------------------------- ------------------------------------------ | time/ | | | fps | 1773 | | iterations | 117 | | time_elapsed | 1080 | | total_timesteps | 1916928 | | train/ | | | approx_kl | 0.0077611385 | | clip_fraction | 0.0792 | | clip_range | 0.2 | | entropy_loss | -3.38 | | explained_variance | 0.931 | | learning_rate | 0.0003 | | loss | -0.0374 | | n_updates | 1160 | | policy_gradient_loss | -0.00399 | | std | 1.31 | | value_loss | 0.00292 | ------------------------------------------ ----------------------------------------- | time/ | | | fps | 1780 | | iterations | 118 | | time_elapsed | 1085 | | total_timesteps | 1933312 | | train/ | | | approx_kl | 0.006831214 | | clip_fraction | 0.0758 | | clip_range | 0.2 | | entropy_loss | -3.4 | | explained_variance | 0.963 | | learning_rate | 0.0003 | | loss | -0.0175 | | n_updates | 1170 | | policy_gradient_loss | -0.00471 | | std | 1.33 | | value_loss | 0.00235 | ----------------------------------------- ----------------------------------------- | time/ | | | fps | 1786 | | iterations | 119 | | time_elapsed | 1091 | | total_timesteps | 1949696 | | train/ | | | approx_kl | 0.006474304 | | clip_fraction | 0.0666 | | clip_range | 0.2 | | entropy_loss | -3.43 | | explained_variance | 0.931 | | learning_rate | 0.0003 | | loss | -0.0318 | | n_updates | 1180 | | policy_gradient_loss | -0.00285 | | std | 1.35 | | value_loss | 0.00699 | ----------------------------------------- Eval num_timesteps=1950000, episode_reward=-35.80 +/- 28.95 Episode length: 2000.00 +/- 0.00 ----------------------------------------- | eval/ | | | mean_ep_length | 2e+03 | | mean_reward | -35.8 | | time/ | | | total_timesteps | 1950000 | | train/ | | | approx_kl | 0.008532442 | | clip_fraction | 0.0746 | | clip_range | 0.2 | | entropy_loss | -3.43 | | explained_variance | 0.958 | | learning_rate | 0.0003 | | loss | -0.00337 | | n_updates | 1190 | | policy_gradient_loss | -0.00376 | | std | 1.34 | | value_loss | 0.0156 | ----------------------------------------- [Diag @ 1,950,000 | n_sheep=4 | success=0%] NEVER_COMPACT 19/20 COMPACT_CANT_DRIVE 1/20 action_mag mean=0.049 p10=0.007 p90=0.044 (0=stopped, 1=full speed) min_flock_radius mean=8.95m best=4.96m (target <5m to compact) min_dog_to_com mean=0.39m best=0.07m (FLEE_DIST=7m) min_com_to_pen mean=14.18m best=9.30m reward/step (mean): progress=-0.0121 alignment=+0.0010 pen_bonus=+0.0005 step_cost=-0.0200 complete=+0.0000 -------------------------------- | time/ | | | fps | 1759 | | iterations | 120 | | time_elapsed | 1117 | | total_timesteps | 1966080 | -------------------------------- ----------------------------------------- | time/ | | | fps | 1766 | | iterations | 121 | | time_elapsed | 1122 | | total_timesteps | 1982464 | | train/ | | | approx_kl | 0.006549825 | | clip_fraction | 0.0665 | | clip_range | 0.2 | | entropy_loss | -3.43 | | explained_variance | 0.966 | | learning_rate | 0.0003 | | loss | -0.0345 | | n_updates | 1200 | | policy_gradient_loss | -0.00349 | | std | 1.34 | | value_loss | 0.00315 | ----------------------------------------- ------------------------------------------ | time/ | | | fps | 1773 | | iterations | 122 | | time_elapsed | 1127 | | total_timesteps | 1998848 | | train/ | | | approx_kl | 0.0062008686 | | clip_fraction | 0.0699 | | clip_range | 0.2 | | entropy_loss | -3.44 | | explained_variance | 0.959 | | learning_rate | 0.0003 | | loss | -0.0512 | | n_updates | 1210 | | policy_gradient_loss | -0.00291 | | std | 1.35 | | value_loss | 0.00544 | ------------------------------------------ Eval num_timesteps=2000000, episode_reward=-45.28 +/- 26.78 Episode length: 2000.00 +/- 0.00 ----------------------------------------- | eval/ | | | mean_ep_length | 2e+03 | | mean_reward | -45.3 | | time/ | | | total_timesteps | 2000000 | | train/ | | | approx_kl | 0.006553275 | | clip_fraction | 0.0739 | | clip_range | 0.2 | | entropy_loss | -3.45 | | explained_variance | 0.924 | | learning_rate | 0.0003 | | loss | -0.0416 | | n_updates | 1220 | | policy_gradient_loss | -0.00427 | | std | 1.36 | | value_loss | 0.0127 | ----------------------------------------- -------------------------------- | time/ | | | fps | 1761 | | iterations | 123 | | time_elapsed | 1144 | | total_timesteps | 2015232 | -------------------------------- ------------------------------------------ | time/ | | | fps | 1767 | | iterations | 124 | | time_elapsed | 1149 | | total_timesteps | 2031616 | | train/ | | | approx_kl | 0.0059226304 | | clip_fraction | 0.0653 | | clip_range | 0.2 | | entropy_loss | -3.46 | | explained_variance | 0.947 | | learning_rate | 0.0003 | | loss | -0.025 | | n_updates | 1230 | | policy_gradient_loss | -0.00273 | | std | 1.36 | | value_loss | 0.00879 | ------------------------------------------ ------------------------------------------ | time/ | | | fps | 1775 | | iterations | 125 | | time_elapsed | 1153 | | total_timesteps | 2048000 | | train/ | | | approx_kl | 0.0076779695 | | clip_fraction | 0.0729 | | clip_range | 0.2 | | entropy_loss | -3.47 | | explained_variance | 0.931 | | learning_rate | 0.0003 | | loss | -0.0382 | | n_updates | 1240 | | policy_gradient_loss | -0.00385 | | std | 1.37 | | value_loss | 0.00692 | ------------------------------------------ Eval num_timesteps=2050000, episode_reward=-44.22 +/- 28.52 Episode length: 2000.00 +/- 0.00 ------------------------------------------ | eval/ | | | mean_ep_length | 2e+03 | | mean_reward | -44.2 | | time/ | | | total_timesteps | 2050000 | | train/ | | | approx_kl | 0.0073502595 | | clip_fraction | 0.0822 | | clip_range | 0.2 | | entropy_loss | -3.49 | | explained_variance | 0.946 | | learning_rate | 0.0003 | | loss | -0.0342 | | n_updates | 1250 | | policy_gradient_loss | -0.00592 | | std | 1.39 | | value_loss | 0.00555 | ------------------------------------------ -------------------------------- | time/ | | | fps | 1764 | | iterations | 126 | | time_elapsed | 1170 | | total_timesteps | 2064384 | -------------------------------- ----------------------------------------- | time/ | | | fps | 1770 | | iterations | 127 | | time_elapsed | 1175 | | total_timesteps | 2080768 | | train/ | | | approx_kl | 0.006628736 | | clip_fraction | 0.0767 | | clip_range | 0.2 | | entropy_loss | -3.51 | | explained_variance | 0.95 | | learning_rate | 0.0003 | | loss | -0.035 | | n_updates | 1260 | | policy_gradient_loss | -0.00457 | | std | 1.4 | | value_loss | 0.00416 | ----------------------------------------- ------------------------------------------ | time/ | | | fps | 1776 | | iterations | 128 | | time_elapsed | 1180 | | total_timesteps | 2097152 | | train/ | | | approx_kl | 0.0068027405 | | clip_fraction | 0.0719 | | clip_range | 0.2 | | entropy_loss | -3.53 | | explained_variance | 0.891 | | learning_rate | 0.0003 | | loss | -0.0391 | | n_updates | 1270 | | policy_gradient_loss | -0.00312 | | std | 1.42 | | value_loss | 0.00492 | ------------------------------------------ Eval num_timesteps=2100000, episode_reward=-39.37 +/- 34.76 Episode length: 2000.00 +/- 0.00 ----------------------------------------- | eval/ | | | mean_ep_length | 2e+03 | | mean_reward | -39.4 | | time/ | | | total_timesteps | 2100000 | | train/ | | | approx_kl | 0.005523986 | | clip_fraction | 0.0604 | | clip_range | 0.2 | | entropy_loss | -3.54 | | explained_variance | 0.938 | | learning_rate | 0.0003 | | loss | -0.0364 | | n_updates | 1280 | | policy_gradient_loss | -0.00281 | | std | 1.42 | | value_loss | 0.015 | ----------------------------------------- [Diag @ 2,100,000 | n_sheep=4 | success=0%] NEVER_COMPACT 20/20 action_mag mean=0.047 p10=0.002 p90=0.041 (0=stopped, 1=full speed) min_flock_radius mean=8.62m best=5.89m (target <5m to compact) min_dog_to_com mean=0.46m best=0.04m (FLEE_DIST=7m) min_com_to_pen mean=14.19m best=7.53m reward/step (mean): progress=-0.0012 alignment=+0.0012 pen_bonus=+0.0010 step_cost=-0.0200 complete=+0.0000 -------------------------------- | time/ | | | fps | 1751 | | iterations | 129 | | time_elapsed | 1206 | | total_timesteps | 2113536 | -------------------------------- ----------------------------------------- | time/ | | | fps | 1756 | | iterations | 130 | | time_elapsed | 1212 | | total_timesteps | 2129920 | | train/ | | | approx_kl | 0.007766474 | | clip_fraction | 0.0823 | | clip_range | 0.2 | | entropy_loss | -3.53 | | explained_variance | 0.96 | | learning_rate | 0.0003 | | loss | -0.0396 | | n_updates | 1290 | | policy_gradient_loss | -0.00492 | | std | 1.41 | | value_loss | 0.00554 | ----------------------------------------- ----------------------------------------- | time/ | | | fps | 1762 | | iterations | 131 | | time_elapsed | 1217 | | total_timesteps | 2146304 | | train/ | | | approx_kl | 0.006704482 | | clip_fraction | 0.0748 | | clip_range | 0.2 | | entropy_loss | -3.53 | | explained_variance | 0.97 | | learning_rate | 0.0003 | | loss | -0.0466 | | n_updates | 1300 | | policy_gradient_loss | -0.00339 | | std | 1.42 | | value_loss | 0.00432 | ----------------------------------------- Eval num_timesteps=2150000, episode_reward=-43.17 +/- 26.95 Episode length: 2000.00 +/- 0.00 ------------------------------------------ | eval/ | | | mean_ep_length | 2e+03 | | mean_reward | -43.2 | | time/ | | | total_timesteps | 2150000 | | train/ | | | approx_kl | 0.0065447316 | | clip_fraction | 0.0751 | | clip_range | 0.2 | | entropy_loss | -3.53 | | explained_variance | 0.888 | | learning_rate | 0.0003 | | loss | -0.0369 | | n_updates | 1310 | | policy_gradient_loss | -0.00369 | | std | 1.41 | | value_loss | 0.0165 | ------------------------------------------ -------------------------------- | time/ | | | fps | 1750 | | iterations | 132 | | time_elapsed | 1235 | | total_timesteps | 2162688 | -------------------------------- ------------------------------------------ | time/ | | | fps | 1755 | | iterations | 133 | | time_elapsed | 1241 | | total_timesteps | 2179072 | | train/ | | | approx_kl | 0.0070872563 | | clip_fraction | 0.075 | | clip_range | 0.2 | | entropy_loss | -3.54 | | explained_variance | 0.954 | | learning_rate | 0.0003 | | loss | -0.0427 | | n_updates | 1320 | | policy_gradient_loss | -0.00406 | | std | 1.42 | | value_loss | 0.00977 | ------------------------------------------ ------------------------------------------ | time/ | | | fps | 1762 | | iterations | 134 | | time_elapsed | 1245 | | total_timesteps | 2195456 | | train/ | | | approx_kl | 0.0073371828 | | clip_fraction | 0.077 | | clip_range | 0.2 | | entropy_loss | -3.55 | | explained_variance | 0.939 | | learning_rate | 0.0003 | | loss | -0.0303 | | n_updates | 1330 | | policy_gradient_loss | -0.00371 | | std | 1.43 | | value_loss | 0.00862 | ------------------------------------------ Eval num_timesteps=2200000, episode_reward=-40.81 +/- 44.39 Episode length: 2000.00 +/- 0.00 ------------------------------------------ | eval/ | | | mean_ep_length | 2e+03 | | mean_reward | -40.8 | | time/ | | | total_timesteps | 2200000 | | train/ | | | approx_kl | 0.0072064474 | | clip_fraction | 0.0714 | | clip_range | 0.2 | | entropy_loss | -3.58 | | explained_variance | 0.951 | | learning_rate | 0.0003 | | loss | -0.0517 | | n_updates | 1340 | | policy_gradient_loss | -0.00405 | | std | 1.45 | | value_loss | 0.00351 | ------------------------------------------ -------------------------------- | time/ | | | fps | 1751 | | iterations | 135 | | time_elapsed | 1262 | | total_timesteps | 2211840 | -------------------------------- ----------------------------------------- | time/ | | | fps | 1758 | | iterations | 136 | | time_elapsed | 1267 | | total_timesteps | 2228224 | | train/ | | | approx_kl | 0.008551812 | | clip_fraction | 0.0911 | | clip_range | 0.2 | | entropy_loss | -3.58 | | explained_variance | 0.929 | | learning_rate | 0.0003 | | loss | -0.0258 | | n_updates | 1350 | | policy_gradient_loss | -0.00599 | | std | 1.45 | | value_loss | 0.0034 | ----------------------------------------- ----------------------------------------- | time/ | | | fps | 1764 | | iterations | 137 | | time_elapsed | 1271 | | total_timesteps | 2244608 | | train/ | | | approx_kl | 0.006960677 | | clip_fraction | 0.0702 | | clip_range | 0.2 | | entropy_loss | -3.59 | | explained_variance | 0.9 | | learning_rate | 0.0003 | | loss | -0.0396 | | n_updates | 1360 | | policy_gradient_loss | -0.00412 | | std | 1.46 | | value_loss | 0.00429 | ----------------------------------------- Eval num_timesteps=2250000, episode_reward=-37.92 +/- 31.68 Episode length: 2000.00 +/- 0.00 ----------------------------------------- | eval/ | | | mean_ep_length | 2e+03 | | mean_reward | -37.9 | | time/ | | | total_timesteps | 2250000 | | train/ | | | approx_kl | 0.005949891 | | clip_fraction | 0.0683 | | clip_range | 0.2 | | entropy_loss | -3.59 | | explained_variance | 0.948 | | learning_rate | 0.0003 | | loss | -0.0381 | | n_updates | 1370 | | policy_gradient_loss | -0.00328 | | std | 1.46 | | value_loss | 0.0113 | ----------------------------------------- [Diag @ 2,250,000 | n_sheep=4 | success=0%] NEVER_COMPACT 19/20 COMPACT_CANT_DRIVE 1/20 action_mag mean=0.068 p10=0.004 p90=0.045 (0=stopped, 1=full speed) min_flock_radius mean=7.87m best=3.57m (target <5m to compact) min_dog_to_com mean=0.45m best=0.15m (FLEE_DIST=7m) min_com_to_pen mean=14.06m best=6.95m reward/step (mean): progress=-0.0035 alignment=+0.0020 pen_bonus=+0.0008 step_cost=-0.0200 complete=+0.0000 -------------------------------- | time/ | | | fps | 1743 | | iterations | 138 | | time_elapsed | 1297 | | total_timesteps | 2260992 | -------------------------------- ------------------------------------------ | time/ | | | fps | 1749 | | iterations | 139 | | time_elapsed | 1301 | | total_timesteps | 2277376 | | train/ | | | approx_kl | 0.0071727796 | | clip_fraction | 0.0784 | | clip_range | 0.2 | | entropy_loss | -3.6 | | explained_variance | 0.943 | | learning_rate | 0.0003 | | loss | -0.0387 | | n_updates | 1380 | | policy_gradient_loss | -0.0042 | | std | 1.46 | | value_loss | 0.0113 | ------------------------------------------ ----------------------------------------- | time/ | | | fps | 1755 | | iterations | 140 | | time_elapsed | 1306 | | total_timesteps | 2293760 | | train/ | | | approx_kl | 0.006800391 | | clip_fraction | 0.0662 | | clip_range | 0.2 | | entropy_loss | -3.59 | | explained_variance | 0.931 | | learning_rate | 0.0003 | | loss | -0.0283 | | n_updates | 1390 | | policy_gradient_loss | -0.00421 | | std | 1.46 | | value_loss | 0.00659 | ----------------------------------------- Eval num_timesteps=2300000, episode_reward=-47.47 +/- 37.24 Episode length: 2000.00 +/- 0.00 ----------------------------------------- | eval/ | | | mean_ep_length | 2e+03 | | mean_reward | -47.5 | | time/ | | | total_timesteps | 2300000 | | train/ | | | approx_kl | 0.008103053 | | clip_fraction | 0.081 | | clip_range | 0.2 | | entropy_loss | -3.59 | | explained_variance | 0.945 | | learning_rate | 0.0003 | | loss | -0.0433 | | n_updates | 1400 | | policy_gradient_loss | -0.00404 | | std | 1.46 | | value_loss | 0.00796 | ----------------------------------------- -------------------------------- | time/ | | | fps | 1745 | | iterations | 141 | | time_elapsed | 1323 | | total_timesteps | 2310144 | -------------------------------- ------------------------------------------ | time/ | | | fps | 1751 | | iterations | 142 | | time_elapsed | 1328 | | total_timesteps | 2326528 | | train/ | | | approx_kl | 0.0061590094 | | clip_fraction | 0.066 | | clip_range | 0.2 | | entropy_loss | -3.61 | | explained_variance | 0.957 | | learning_rate | 0.0003 | | loss | -0.0436 | | n_updates | 1410 | | policy_gradient_loss | -0.00287 | | std | 1.47 | | value_loss | 0.0102 | ------------------------------------------ ------------------------------------------ | time/ | | | fps | 1757 | | iterations | 143 | | time_elapsed | 1332 | | total_timesteps | 2342912 | | train/ | | | approx_kl | 0.0070403973 | | clip_fraction | 0.0733 | | clip_range | 0.2 | | entropy_loss | -3.62 | | explained_variance | 0.863 | | learning_rate | 0.0003 | | loss | -0.0356 | | n_updates | 1420 | | policy_gradient_loss | -0.00525 | | std | 1.48 | | value_loss | 0.0103 | ------------------------------------------ Eval num_timesteps=2350000, episode_reward=-47.95 +/- 27.60 Episode length: 2000.00 +/- 0.00 ----------------------------------------- | eval/ | | | mean_ep_length | 2e+03 | | mean_reward | -48 | | time/ | | | total_timesteps | 2350000 | | train/ | | | approx_kl | 0.007505033 | | clip_fraction | 0.0729 | | clip_range | 0.2 | | entropy_loss | -3.64 | | explained_variance | 0.94 | | learning_rate | 0.0003 | | loss | -0.0473 | | n_updates | 1430 | | policy_gradient_loss | -0.00385 | | std | 1.5 | | value_loss | 0.00449 | ----------------------------------------- -------------------------------- | time/ | | | fps | 1747 | | iterations | 144 | | time_elapsed | 1350 | | total_timesteps | 2359296 | -------------------------------- ---------------------------------------- | time/ | | | fps | 1752 | | iterations | 145 | | time_elapsed | 1355 | | total_timesteps | 2375680 | | train/ | | | approx_kl | 0.00724002 | | clip_fraction | 0.0739 | | clip_range | 0.2 | | entropy_loss | -3.65 | | explained_variance | 0.948 | | learning_rate | 0.0003 | | loss | -0.0419 | | n_updates | 1440 | | policy_gradient_loss | -0.00426 | | std | 1.5 | | value_loss | 0.00886 | ---------------------------------------- ----------------------------------------- | time/ | | | fps | 1758 | | iterations | 146 | | time_elapsed | 1360 | | total_timesteps | 2392064 | | train/ | | | approx_kl | 0.007578165 | | clip_fraction | 0.0713 | | clip_range | 0.2 | | entropy_loss | -3.64 | | explained_variance | 0.859 | | learning_rate | 0.0003 | | loss | -0.0427 | | n_updates | 1450 | | policy_gradient_loss | -0.0049 | | std | 1.49 | | value_loss | 0.00429 | ----------------------------------------- Eval num_timesteps=2400000, episode_reward=-47.88 +/- 34.39 Episode length: 2000.00 +/- 0.00 ----------------------------------------- | eval/ | | | mean_ep_length | 2e+03 | | mean_reward | -47.9 | | time/ | | | total_timesteps | 2400000 | | train/ | | | approx_kl | 0.006707498 | | clip_fraction | 0.0692 | | clip_range | 0.2 | | entropy_loss | -3.65 | | explained_variance | 0.861 | | learning_rate | 0.0003 | | loss | -0.0426 | | n_updates | 1460 | | policy_gradient_loss | -0.00411 | | std | 1.5 | | value_loss | 0.00639 | ----------------------------------------- [Diag @ 2,400,000 | n_sheep=4 | success=0%] NEVER_COMPACT 19/20 COMPACT_CANT_DRIVE 1/20 action_mag mean=0.052 p10=0.005 p90=0.045 (0=stopped, 1=full speed) min_flock_radius mean=8.79m best=3.32m (target <5m to compact) min_dog_to_com mean=0.45m best=0.20m (FLEE_DIST=7m) min_com_to_pen mean=13.96m best=9.02m reward/step (mean): progress=-0.0047 alignment=+0.0013 pen_bonus=+0.0005 step_cost=-0.0200 complete=+0.0000 -------------------------------- | time/ | | | fps | 1737 | | iterations | 147 | | time_elapsed | 1386 | | total_timesteps | 2408448 | -------------------------------- Training complete. Artefacts saved to runs/ppo_fix_check2/