jcreek/CosmicClash
1
0
Fork 0
mirror of https://github.com/jcreek/CosmicClash.git synced 2026-09-13 07:52:28 +00:00
Code Issues Packages Projects Releases Wiki Activity

fix(training): make uprightness a multiplier on the forward-approach reward

Browse Source 
Rounds 2 and 3 showed that tuning grounded_upright_reward's magnitude only
slides along a tradeoff instead of resolving it: at 0.015 upright_fraction
climbed to 0.331 while goal_rate sagged to 0.542 (then farmed outright at
0.696/0.366), and at 0.004 goal_rate climbed 0.569->0.604 while
upright_fraction went flat at ~0.26. An additive uprightness bonus is an
alternative to playing well, so the policy just picks whichever is cheaper
and no magnitude buys both behaviours.

Change the mechanism rather than the number: grounded_upright_reward drops
to 0, and uprightness becomes a multiplier inside the nose-led approach
term, which already requires moving forward at the ball. Parked-and-upright
and fast-but-sideways now both pay zero; only upright, forward, nose-on to
the ball pays full. forward-velocity-to-ball rises 0.06 -> 0.15 to offset
the ~2-3x expected-value cut from the new factor, and ground-tilt-penalty
drops 0.05 -> 0.02 now that uprightness is paid positively during play.
Delete the three blocked attempts and reset state to restart from the
Stage-3 foundation.
This commit is contained in:
Josh Creek
2026-08-12 13:04:20 +01:00
parent 5260124274
commit 4f3cf56e28
12 changed files with 62 additions and 267 deletions
Download Patch File Download Diff File Expand all files Collapse all files
training/generation5_state.json
+3 -257
View File
@@ -1,260 +1,6 @@
{
"stage_index": 0,
"attempt": 2,
"status": "blocked",
"log": [
{
"stage_index": 0,
"stage_number": 4,
"stage_name": "handling",
"experiment": "20260811-1606-gen5-s4-handling",
"attempt": 0,
"telemetry_tail": {
"rollout/air_touch_fraction": 0.0004399999901652336,
"rollout/airborne_fraction": 0.44587357181310655,
"rollout/ep_len_mean": 139.7144799041748,
"rollout/ep_rew_mean": -5.688299763679504,
"rollout/forward_motion_fraction": 0.21974961887300015,
"rollout/goal_rate": 0.5693600015640259,
"rollout/mean_altitude": 4.424891119480133,
"rollout/productive_air_touch_fraction": 7.999999821186065e-05,
"rollout/upright_fraction": 0.2615443330407143,
"rollout/vertical_thrust_mean": 0.08650099822867197
},
"telemetry_failures": [
"rollout/goal_rate=0.5694 < 0.8000",
"rollout/upright_fraction=0.2615 < 0.4500",
"rollout/forward_motion_fraction=0.2197 < 0.2500"
],
"evaluation_goal_failures": [
"easy.json: goal_rate=0.770 < 0.800"
],
"side_balance_failures": [],
"eval": {
"timestamp": "2026-08-11T21:16:10+00:00",
"model_a": "/home/jcreek/ai-training/CosmicClash/Game/bots/20260811-1606-gen5-s4-handling.json",
"model_b": "/home/jcreek/ai-training/CosmicClash/Game/bots/promoted/easy.json",
"seed": 1,
"episodes": 100,
"wins_a": 47,
"wins_b": 30,
"draws": 23,
"side_results": {
"a_team_0": {
"wins_a": 22,
"wins_b": 16,
"draws": 12
},
"a_team_1": {
"wins_a": 25,
"wins_b": 14,
"draws": 11
}
},
"physical_team_wins": {
"team_0": 36,
"team_1": 41
},
"win_rate_a": 0.47
},
"evals": [
{
"timestamp": "2026-08-11T21:16:10+00:00",
"model_a": "/home/jcreek/ai-training/CosmicClash/Game/bots/20260811-1606-gen5-s4-handling.json",
"model_b": "/home/jcreek/ai-training/CosmicClash/Game/bots/promoted/easy.json",
"seed": 1,
"episodes": 100,
"wins_a": 47,
"wins_b": 30,
"draws": 23,
"side_results": {
"a_team_0": {
"wins_a": 22,
"wins_b": 16,
"draws": 12
},
"a_team_1": {
"wins_a": 25,
"wins_b": 14,
"draws": 11
}
},
"physical_team_wins": {
"team_0": 36,
"team_1": 41
},
"win_rate_a": 0.47
}
],
"decision": "fail"
},
{
"stage_index": 0,
"stage_number": 4,
"stage_name": "handling",
"experiment": "20260811-2216-gen5-s4-handling-retry1",
"attempt": 1,
"telemetry_tail": {
"rollout/air_touch_fraction": 0.00019999999552965163,
"rollout/airborne_fraction": 0.446684571146965,
"rollout/ep_len_mean": 135.95963993835448,
"rollout/ep_rew_mean": -4.255322490215302,
"rollout/forward_motion_fraction": 0.21915800005197525,
"rollout/goal_rate": 0.5982800037264824,
"rollout/mean_altitude": 4.420792480945587,
"rollout/productive_air_touch_fraction": 3.999999910593033e-05,
"rollout/upright_fraction": 0.2567908099591732,
"rollout/vertical_thrust_mean": 0.1063969987387536
},
"telemetry_failures": [
"rollout/goal_rate=0.5983 < 0.8000",
"rollout/upright_fraction=0.2568 < 0.4500",
"rollout/forward_motion_fraction=0.2192 < 0.2500"
],
"evaluation_goal_failures": [],
"side_balance_failures": [],
"eval": {
"timestamp": "2026-08-12T03:24:21+00:00",
"model_a": "/home/jcreek/ai-training/CosmicClash/Game/bots/20260811-2216-gen5-s4-handling-retry1.json",
"model_b": "/home/jcreek/ai-training/CosmicClash/Game/bots/promoted/easy.json",
"seed": 1,
"episodes": 100,
"wins_a": 53,
"wins_b": 29,
"draws": 18,
"side_results": {
"a_team_0": {
"wins_a": 22,
"wins_b": 17,
"draws": 11
},
"a_team_1": {
"wins_a": 31,
"wins_b": 12,
"draws": 7
}
},
"physical_team_wins": {
"team_0": 34,
"team_1": 48
},
"win_rate_a": 0.53
},
"evals": [
{
"timestamp": "2026-08-12T03:24:21+00:00",
"model_a": "/home/jcreek/ai-training/CosmicClash/Game/bots/20260811-2216-gen5-s4-handling-retry1.json",
"model_b": "/home/jcreek/ai-training/CosmicClash/Game/bots/promoted/easy.json",
"seed": 1,
"episodes": 100,
"wins_a": 53,
"wins_b": 29,
"draws": 18,
"side_results": {
"a_team_0": {
"wins_a": 22,
"wins_b": 17,
"draws": 11
},
"a_team_1": {
"wins_a": 31,
"wins_b": 12,
"draws": 7
}
},
"physical_team_wins": {
"team_0": 34,
"team_1": 48
},
"win_rate_a": 0.53
}
],
"decision": "fail"
},
{
"stage_index": 0,
"stage_number": 4,
"stage_name": "handling",
"experiment": "20260812-0424-gen5-s4-handling-retry2",
"attempt": 2,
"telemetry_tail": {
"rollout/air_touch_fraction": 0.00025999999418854715,
"rollout/airborne_fraction": 0.44747342938184737,
"rollout/ep_len_mean": 133.66904002380372,
"rollout/ep_rew_mean": -2.8346366541385652,
"rollout/forward_motion_fraction": 0.22339595240354537,
"rollout/goal_rate": 0.6036400033831596,
"rollout/mean_altitude": 4.436961054801941,
"rollout/productive_air_touch_fraction": 7.999999821186065e-05,
"rollout/upright_fraction": 0.2713429524004459,
"rollout/vertical_thrust_mean": 0.05840199833672028
},
"telemetry_failures": [
"rollout/goal_rate=0.6036 < 0.8000",
"rollout/upright_fraction=0.2713 < 0.4500",
"rollout/forward_motion_fraction=0.2234 < 0.2500"
],
"evaluation_goal_failures": [
"easy.json: goal_rate=0.660 < 0.800"
],
"side_balance_failures": [],
"eval": {
"timestamp": "2026-08-12T09:31:39+00:00",
"model_a": "/home/jcreek/ai-training/CosmicClash/Game/bots/20260812-0424-gen5-s4-handling-retry2.json",
"model_b": "/home/jcreek/ai-training/CosmicClash/Game/bots/promoted/easy.json",
"seed": 1,
"episodes": 100,
"wins_a": 37,
"wins_b": 29,
"draws": 34,
"side_results": {
"a_team_0": {
"wins_a": 18,
"wins_b": 20,
"draws": 12
},
"a_team_1": {
"wins_a": 19,
"wins_b": 9,
"draws": 22
}
},
"physical_team_wins": {
"team_0": 27,
"team_1": 39
},
"win_rate_a": 0.37
},
"evals": [
{
"timestamp": "2026-08-12T09:31:39+00:00",
"model_a": "/home/jcreek/ai-training/CosmicClash/Game/bots/20260812-0424-gen5-s4-handling-retry2.json",
"model_b": "/home/jcreek/ai-training/CosmicClash/Game/bots/promoted/easy.json",
"seed": 1,
"episodes": 100,
"wins_a": 37,
"wins_b": 29,
"draws": 34,
"side_results": {
"a_team_0": {
"wins_a": 18,
"wins_b": 20,
"draws": 12
},
"a_team_1": {
"wins_a": 19,
"wins_b": 9,
"draws": 22
}
},
"physical_team_wins": {
"team_0": 27,
"team_1": 39
},
"win_rate_a": 0.37
}
],
"decision": "fail"
}
]
"attempt": 0,
"status": "in_progress",
"log": []
}