Created on wrong repo, see SirStone/tank-royale
This issue is blocked by:
- #50 (Tier 1 SharedTestRunner) — tests must be green before tagging
- #51 (CI workflow) — CI must be green on all matrix targets before publishing
Do not start this…
SAC training module complete — SAC-v2 with manual backprop, truncated BPTT (ponytail: full BPTT if gradient quality matters), twin critics, auto-alpha, soft target update. 11/11 training tests…
Weight persistence module complete — zipped .npy via std/zipfiles, atomic save (temp+rename), checkpoint includes Adam states with optional load. 12/12 tests pass. Full suite 69+ tests all green.…
Replay buffer module complete — sequential ring buffer, battle-boundary episodes, burn-in/train split. 10/10 tests pass.
Reward module complete — all event types, Welford running normalization, cold-start safe. All tests pass.
Action mapping module complete — speed-aware turn, asymmetric accel [-2,+1], fire threshold. 19/19 tests pass.
State vector module complete — 35-dim normalized tensor, bullet tracking, scan staleness. 5/5 tests pass.
LSTM network module complete — actor + dual critics with proper LSTM cells, squashed Gaussian policy, deterministic mode. 9/9 tests pass.