jalf/TIR_PROJ
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

Approach refinement

Browse Source
This commit is contained in:
Johnny Fernandes
2026-04-26 02:02:25 +01:00
parent 61f8a7db15
commit 287743709a
2 changed files with 7 additions and 8 deletions
Download Patch File Download Diff File Expand all files Collapse all files
training/config.json
+2 -2
View File
@@ -5,8 +5,8 @@
"W_COMPLETE": 100.0, "W_COMPLETE": 100.0,
"W_STEP_COST": 0.02, "W_STEP_COST": 0.02,
"W_COMPACT": 0.0, "W_COMPACT": 0.0,
"W_WALL_TOUCH": 0.15, "W_WALL_TOUCH": 0.04,
"WALL_TOUCH_BUFFER": 0.8, "WALL_TOUCH_BUFFER": 0.3,
"ALIGN_SHAPE": "standoff", "ALIGN_SHAPE": "standoff",
"ALIGN_GATED": true, "ALIGN_GATED": true,
"ENTRY_AWARE": false, "ENTRY_AWARE": false,
training/herding_env.py
+5 -6
View File
@@ -61,12 +61,11 @@ class HerdingEnv(gym.Env):
W_COMPLETE = 100.0 # all sheep penned W_COMPLETE = 100.0 # all sheep penned
W_STEP_COST = 0.02 # time penalty — strong enough to punish doing nothing W_STEP_COST = 0.02 # time penalty — strong enough to punish doing nothing
W_COMPACT = 0.0 # reward for flock-radius reduction (off by default) W_COMPACT = 0.0 # reward for flock-radius reduction (off by default)
W_WALL_TOUCH = 0.15 # per-sheep max penalty at wall surface. Linear ramp W_WALL_TOUCH = 0.04 # per-sheep max penalty at wall surface. Linear ramp
# within WALL_TOUCH_BUFFER gives the RL agent a gradient # within WALL_TOUCH_BUFFER nudges the agent to avoid
# signal to avoid pinning sheep against pen walls. # pinning sheep against pen walls. 0.04 ≈ 2× step_cost
# 0.15 ≈ 7.5× step_cost — strong enough to shape behavior # — noticeable but never dominates progress reward.
# without overwhelming progress reward. WALL_TOUCH_BUFFER = 0.3 # metres from wall where penalty starts ramping
WALL_TOUCH_BUFFER = 0.8 # metres from wall where penalty starts ramping
ALIGN_SHAPE = "standoff" # "standoff" (peaks at IDEAL) | "near" (peaks at 0) ALIGN_SHAPE = "standoff" # "standoff" (peaks at IDEAL) | "near" (peaks at 0)
ALIGN_GATED = True # gate alignment on action magnitude ALIGN_GATED = True # gate alignment on action magnitude
ENTRY_AWARE = False # When True, targets PEN_ENTRY (entrance face) instead ENTRY_AWARE = False # When True, targets PEN_ENTRY (entrance face) instead