Forge GearMesh expert (PPO)
PPO policy for Isaac-Forge-GearMesh-Direct-v0 (Isaac Lab FORGE suite),
trained with rl_games. Final return 740.3; measured success 12/12 in scripted video evaluation; ~99% over ~1,004 collection episodes.
What is in the file
rl_games checkpoints hold more than the policy: the actor-critic weights, the optimizer state,
and — importantly for inference — the running observation normalisation statistics. That is why a
small MLP policy takes 204 MB. Load it through the Isaac Lab player rather than torch.load alone,
so the normaliser is restored with the weights:
./isaaclab.sh -p scripts/reinforcement_learning/rl_games/play.py \
--task Isaac-Forge-GearMesh-Direct-v0 --checkpoint /path/to/forge_gearmesh_2026-08-01.pth --num_envs 1 --headless
Training
Stock Isaac Lab rl_games PPO, unchanged hyper-parameters except the epoch count:
128 environments, seed 0, 300 epochs (the shipped config stops at 200, which cuts the run while
the return is still climbing). Snapshots exist at epochs 100 and 200; this repo ships epoch 300.
Related
- Demonstrations produced with this policy:
angledusgar/forge-v1 - Its failures:
angledusgar/forge-failure-v1