Compare commits

...

2 Commits

Author SHA1 Message Date
CosmicClash Training Bot d06a67ade6 chore(training): generation 5 progress after 20260824-1052-gen5-s5-intercepts 2026-08-25 08:35:28 +01:00
CosmicClash Training Bot 6d94366693 chore(training): Add 20260824-1052-gen5-s5-intercepts checkpoints, logs, and exported policy 2026-08-25 08:32:51 +01:00
5 changed files with 116 additions and 7 deletions
File diff suppressed because one or more lines are too long
+27
View File
@@ -1089,5 +1089,32 @@
"team_1": 46
},
"win_rate_a": 0.36
},
{
"timestamp": "2026-08-25T07:35:28+00:00",
"model_a": "/home/jcreek/ai-training/CosmicClash/Game/bots/20260824-1052-gen5-s5-intercepts.json",
"model_b": "/home/jcreek/ai-training/CosmicClash/Game/bots/20260816-2126-gen5-s4-handling-retry2.json",
"seed": 1,
"episodes": 100,
"wins_a": 45,
"wins_b": 32,
"draws": 23,
"side_results": {
"a_team_0": {
"wins_a": 22,
"wins_b": 16,
"draws": 12
},
"a_team_1": {
"wins_a": 23,
"wins_b": 16,
"draws": 11
}
},
"physical_team_wins": {
"team_0": 38,
"team_1": 39
},
"win_rate_a": 0.45
}
]
+86 -5
View File
@@ -1,6 +1,6 @@
{
"stage_index": 1,
"attempt": 0,
"attempt": 1,
"status": "in_progress",
"log": [
{
@@ -754,11 +754,92 @@
}
],
"decision": "fail"
},
{
"stage_index": 1,
"stage_number": 5,
"stage_name": "intercepts",
"experiment": "20260824-1052-gen5-s5-intercepts",
"attempt": 0,
"telemetry_tail": {
"rollout/air_touch_fraction": 0.00025999999418854715,
"rollout/airborne_fraction": 0.3494572382569313,
"rollout/ball_above_air_touch_fraction": 0.44360085755586626,
"rollout/ball_mean_altitude": 4.1907234663963315,
"rollout/ball_peak_altitude": 5.405460600852966,
"rollout/ep_len_mean": 109.09664013671875,
"rollout/ep_rew_mean": 19.020949184417724,
"rollout/forward_motion_fraction": 0.50440152323246,
"rollout/goal_rate": 0.7694799978733062,
"rollout/grounded_upright_fraction": 0.1997999999076128,
"rollout/mean_altitude": 3.2880454597473143,
"rollout/productive_air_touch_episode_fraction": 3.999999910593033e-05,
"rollout/productive_air_touch_fraction": 3.999999910593033e-05,
"rollout/upright_fraction": 0.7768516182899475,
"rollout/vertical_thrust_mean": 0.09507499811029993
},
"telemetry_failures": [
"rollout/productive_air_touch_episode_fraction=0.0000 < 0.0200"
],
"evaluation_goal_failures": [],
"side_balance_failures": [],
"eval": {
"timestamp": "2026-08-25T07:35:28+00:00",
"model_a": "/home/jcreek/ai-training/CosmicClash/Game/bots/20260824-1052-gen5-s5-intercepts.json",
"model_b": "/home/jcreek/ai-training/CosmicClash/Game/bots/20260816-2126-gen5-s4-handling-retry2.json",
"seed": 1,
"episodes": 100,
"wins_a": 45,
"wins_b": 32,
"draws": 23,
"side_results": {
"a_team_0": {
"wins_a": 22,
"wins_b": 16,
"draws": 12
},
"a_team_1": {
"wins_a": 23,
"wins_b": 16,
"draws": 11
}
},
"physical_team_wins": {
"team_0": 38,
"team_1": 39
},
"win_rate_a": 0.45
},
"evals": [
{
"timestamp": "2026-08-25T07:35:28+00:00",
"model_a": "/home/jcreek/ai-training/CosmicClash/Game/bots/20260824-1052-gen5-s5-intercepts.json",
"model_b": "/home/jcreek/ai-training/CosmicClash/Game/bots/20260816-2126-gen5-s4-handling-retry2.json",
"seed": 1,
"episodes": 100,
"wins_a": 45,
"wins_b": 32,
"draws": 23,
"side_results": {
"a_team_0": {
"wins_a": 22,
"wins_b": 16,
"draws": 12
},
"a_team_1": {
"wins_a": 23,
"wins_b": 16,
"draws": 11
}
},
"physical_team_wins": {
"team_0": 38,
"team_1": 39
},
"win_rate_a": 0.45
}
],
"resume_override": {
"stage_index": 1,
"experiment": "20260823-1734-gen5-s5-intercepts-retry2",
"reason": "Round 10 (2026-08-24): Stage 5 reopened after nine attempts blocked on a gate that could not register the behaviour it measured. productive_air_touch_fraction divided by total touches (so Stage 4's ground game diluted it), AIR_TOUCH_HEIGHT 5.0 was ~2x the measured mean episode ball peak of 2.4m, and the aggregate entropy controller hid a collapsed thrust_y head that left the policy commanding 0.03 vertical thrust against the 0.408 needed to hover. Metric, bar + drill band, and per-head exploration floor all changed. Unlike Round 9 this DOES change the reward function -- AIR_TOUCH_HEIGHT also gates air_touch_bonus_weight's payout -- but the changed term has never once fired (productive_air_touch_fraction exactly 0.0 for nine attempts), so there is no learned value estimate attached to it, while retry2's ground handling and scoring are untouched and are what is being kept."
"decision": "fail"
}
]
}