Hugging Face's logo

Homalozoa
/
spirule-policies

Reinforcement Learning
robotics
mujoco
Model card Files
xet
 Community

You need to agree to share your Hugging Face information to access this model

This repository is publicly accessible, but you have to accept the conditions to access its files and content.

Log in or Sign Up to review the conditions and access this model content.

spirule policies

PPO policies (rsl_rl, mjlab) for SpiRobs, trained with spirule on MuJoCo Warp, 2026-10-27. Backup of linkp6 /data0/zhanchi.wang/spirule/logs/rsl_rl (last checkpoint of every run) and the target libraries the runs were trained with.

  • checkpoints/<experiment>/<run>/model_<iter>.pt: actor-critic weights and optimizer state; params/env.yaml and params/agent.yaml next to it are the exact training configs.
  • libraries/spirobs_<N>cable/library.npz, library_size.npz: reachable targets, paths and fast points (scripts/make_libraries.py); copy into src/spirule/assets/spirobs_<N>cable/ to train or evaluate.
  • eval.jsonl: every evaluation run (scripts/eval_task.py).
  • SHA256SUMS: checksums of the checkpoints and libraries.

Experiment name -> task: spirule_<kind>_<N>cable[_<variant>] is Spirule-<Kind>-<N>Cable[-<Variant>]; _size is the 1 cm-1 m policy; _sizert the same with motor and camera timing randomized in real seconds (track fine-tuned from _size model_2999).

Main policies:

Task Checkpoint Result (randomized robots)
Reach-2Cable-Size spirule_reach_2cable_size/2026-10-27_16-35-00/model_1499.pt 0.5-0.75 % of length
Reach-3Cable-Size spirule_reach_3cable_size/2026-10-27_19-53-25/model_1499.pt 0.75-1.1 %
Reach-4Cable-Size spirule_reach_4cable_size/2026-10-27_19-54-22/model_1499.pt 0.7-1.2 %
Reach-6Cable-Size spirule_reach_6cable_size/2026-10-27_19-59-18/model_1499.pt 0.6-0.9 %
Track-2Cable-Size spirule_track_2cable_size/2026-10-27_16-35-00/model_2999.pt 2.9-3.7 %
Strike-2Cable-Size spirule_strike_2cable_size/2026-10-27_16-35-00/model_2499.pt 97-100 % hits
Reach-2Cable-SizeRT spirule_reach_2cable_sizert/2026-10-28_05-01-19/model_1999.pt 0.7-0.9 % from 4.6 cm up; 1 cm median 2.3 %
Track-2Cable-SizeRT spirule_track_2cable_sizert/2026-10-28_08-09-18/model_4998.pt 3.3-3.6 % from 46.4 cm up; 10 % at 10 cm
Reach-2Cable-Inc spirule_reach_2cable_inc/2026-10-27_15-32-22/model_599.pt 2.8 mm mean at 46.4 cm
Track-2Cable-Sharp spirule_track_2cable_sharp/2026-10-27_15-32-22/model_1499.pt 13.5 mm mean at 46.4 cm

The other runs are the precision-study variants and the first single-size policies; see the spirule README for what each changes.

python scripts/eval_task.py Spirule-Reach-2Cable-Size \
  --checkpoint checkpoints/spirule_reach_2cable_size/2026-10-27_16-35-00/model_1499.pt --length-m 0.1
Downloads last month

-

Downloads are not tracked for this model. How to track
Video Preview
Reinforcement Learning
loading