OpenRAL/rskill-act-aloha-aloha_insertion-fp32

rskill-act-aloha-insertion

OpenRAL rSkill (custom example) — ACT (Action Chunking Transformer) finetuned on the ALOHA bimanual peg-insertion task, packaged for OpenRAL.

This package wraps lerobot/act_aloha_sim_insertion_human with a rskill.yaml manifest that adds capability checking, license surfacing, latency budgets, and local registry integration. It does not copy model weights.

It is the harder sibling of rskill-act-aloha (cube transfer) and demonstrates how a single packaging format covers multiple task-specific checkpoints from the same paper. The runnable demo lives at scenes/benchmark/aloha_insertion.yaml and is wired into the top-level just sim-custom recipe.

Upstream model

Field	Value
Source repo	lerobot/act_aloha_sim_insertion_human
Architecture	Action Chunking Transformer (~52M params, chunk=100)
Task	gym-aloha AlohaInsertion-v0 (bimanual peg-in-socket)
License	MIT
Paper	Zhao et al., 2023 — Learning Fine-Grained Bimanual Manipulation with Low-Cost Hardware (arXiv 2304.13705)

Why no eval/ block?

This skill is shipped as a custom-example package, not as a reproduced benchmark entry. The paper's headline number for sim ALOHA insertion is markedly lower than the cube-transfer figure (the task is harder and the upstream protocol uses different camera intrinsics). We deliberately omit eval/ rather than copy paper numbers without an internal reproduction; per CLAUDE.md §6.4 that omission must be documented — this section is that documentation. Add eval/aloha_insertion.json once a local reproduction lands.

Supported robots

Robot	Embodiment tag	Status	Notes
ALOHA bimanual (Trossen) — gym-aloha MuJoCo	aloha, lerobot	✓ sim	14-DoF (2 × 7-DoF arms with parallel grippers); MuJoCo MJX AlohaInsertion-v0.

Same physical embodiment as the act-aloha sibling (cube transfer); the only difference is the task contact dynamics — peg insertion is harder than cube pick-and-place.

Sensors required

Key	Modality	Resolution	Format
observation.images.top	RGB camera	640 × 480	float32
observation.state	proprioception	(14,)	float32 (2 × 7-DoF joint positions)

Single top-down RGB stream like the cube-transfer sibling — the checkpoint does not consume wrist or third-person views.

Manifest summary

Field	Value
name	OpenRAL/rskill-act-aloha-insertion
version	0.1.0
license	mit
role	s1
embodiment_tags	aloha, lerobot
runtime / quantization.dtype	pytorch / fp32
weights_uri	hf://lerobot/act_aloha_sim_insertion_human
chunk_size	100
commercial_use_allowed	true

Full schema: openral_core.schemas.RSkillManifest.

Run it

just sim-custom

…which is equivalent to:

MUJOCO_GL=egl uv run --group sim openral sim run \
  --config scenes/benchmark/aloha_insertion.yaml \
  --save-video example_videos

License

This rSkill package (rskill.yaml, README.md) is MIT to match the upstream weights. Commercial use is allowed (commercial_use_allowed: true).