Issues

• SpeedWrapper can return a result dict missing obs, terminated, truncated, and info

  bug
  UHUMALAB/PPAGE#68 · Filter by author ProValarous opened on Jul 25, 2026

• DQN save and load does not persist enough state to safely resume training

  enhancement
  UHUMALAB/PPAGE#66 · Filter by author ProValarous opened on Jul 25, 2026

• Observation, reward, and action registries silently overwrite on duplicate name

  bug
  UHUMALAB/PPAGE#64 · Filter by author ProValarous opened on Jul 25, 2026

• --seed means different things in render.py versus demo_movement.py

  bug
  UHUMALAB/PPAGE#63 · Filter by author ProValarous opened on Jul 25, 2026

• Scripts do not guarantee env.close() runs on exception, can leak the pygame window

  bug
  UHUMALAB/PPAGE#62 · Filter by author ProValarous opened on Jul 25, 2026

• observations.yaml per_agent and rewards.yaml applies_to and normalization are silently ignored

  bug
  UHUMALAB/PPAGE#61 · Filter by author ProValarous opened on Jul 25, 2026

• sweep.py clobbers checkpoint and curve output across sweep iterations

  bug
  UHUMALAB/PPAGE#60 · Filter by author ProValarous opened on Jul 25, 2026

• evaluate.py silently evaluates A2C checkpoints with exploration still on

  bug
  UHUMALAB/PPAGE#59 · Filter by author ProValarous opened on Jul 25, 2026

• ActorCritic always samples stochastically, even during evaluation

  bug
  UHUMALAB/PPAGE#58 · Filter by author ProValarous opened on Jul 25, 2026

• A2C zeros the value bootstrap on truncation, not just true termination

  bug
  UHUMALAB/PPAGE#57 · Filter by author ProValarous opened on Jul 25, 2026