SirStone/SirRoboGarage
1
0
Fork 0
Code Issues 28 Pull Requests Actions Projects Releases Wiki Activity
74 Commits 5 Branches 0 Tags
b7b10af8119ad29b9a92d9fa091da090d60c7db6
Commit Graph

7 Commits

Author SHA1 Message Date
SirStone 0d35646dc9 feat(PPO_Bot): deterministic eval + fix logStd warm-start 
- actorForward: deterministic param, uses mean-only when PPOB_EVAL_ONLY=1
  (eval was adding unit Gaussian noise to every action — unreliable scores)
- warm_start.py: log_std initialized to -1.0 (std≈0.37) instead of copying
  snapshot values (were 2.27-4.68 → std 9-108, completely drowning signal)
- training.env: LOG_STD_CEILING 0.0→-0.5 (cap exploration at std≈0.6)
 2026-08-20 15:18:07 +02:00
SirStone 12624d3069 feat(PPO_Bot): bot-relative bullets + scan staleness (STATE_DIM=57) 
- Bullet state (indices 44-55): enemy-relative → bot-relative frame
  (bot needs threat vectors to itself for dodging, not to enemy)
- New index 56: scan staleness = min(ticksSinceLastScan / 30, 1.0)
  (gives policy a confidence signal for enemy data freshness)
- warm_start.py updated: 44→57 dim expansion, TARGET_DIM variable
- Tests updated for new state layout

Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
 2026-08-20 14:27:28 +02:00
SirStone 6ad51148f4 fix(PPO_Bot): SIGSEGV crash fixes + static buffers for thread safety 
- bullets: seq[InFlightBullet] → array[4, InFlightBullet] + bulletCount
  (eliminates cross-thread heap realloc under ORC)
- hasFired: edge-triggered (cleared after state build, not level-triggered)
- round_counter parseInt: wrapped for empty/torn file → 0
- Static SVG + intent buffers to kill cross-thread heap realloc
- Tick-local alive/bulletData also fixed arrays

Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
 2026-08-20 14:23:08 +02:00
SirStone 766b9e03ee feat(PPO_Bot): enemy-centered action space + reward shaping for 100% vs Target 
- actions.nim: goto/aimTo coordinates now offset from enemy position
  (enemyX + tanh(raw) * scale) instead of absolute arena coords
  (sigmoid(raw) * arenaSize). Initial random policy defaults to
  approaching and aiming at enemy.
- training.nim: added dense reward shaping (distance closeness +
  gun bearing) to computeTickReward, doubled round reward scaling.
- PPO_Bot.nim: passes enemy position to mapActions, computes
  gun-to-enemy bearing for reward shaping.

Result: 100/100 win rate vs Target with frozen weights.

Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
 2026-08-18 17:40:59 +02:00
SirStone cdde60d79f feat(PPO_Bot): command abstraction layer — goto/aimTo controllers (#24) 
- Add gotoTick/aimToTick controller functions (#25)
- Update network dims: actor 5→6, state 42→44 (#26)
- Rewrite mapActions for 6-dim command space (#27)
- Delete stale weight files (shape mismatch)
- Fix existing tests for new signatures

Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
 2026-08-17 19:20:40 +02:00
SirStone eadd177d3b feat(PPO_Bot): reward + trajectory + GAE + PPO training (#16) 
Manual-backprop PPO with Adam: TrajectoryBuffer, computeGAE, ppoUpdate
(4 epochs, minibatch 64, clip 0.2, grad norm 0.5). Reward helpers
computeTickReward/computeRoundReward. Bot wired: tick transitions
collected in run loop, ppoUpdate called on onRoundEnded. Fix: add
arraymancer import to PPO_Bot.nim so Tensor resolves at top level.

Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
 2026-08-16 15:27:15 +02:00
SirStone aa4bc77068 feat(PPO_Bot): network forward pass + action mapping (#14) 
Two-hidden-layer MLP actor-critic (42→64→64→5/1) with stochastic
actorForward, logStd floor at -3, and BotAction mapper wired into
the run() loop. Assert-based test suite covers shapes, finiteness,
logStd collapse, and all action range bounds.

Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
 2026-08-16 15:10:46 +02:00