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 examples/sim/custom_act_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.

Run it

just sim-custom

…which is equivalent to:

MUJOCO_GL=egl uv run --group sim ral sim run \
  --config examples/sim/custom_act_aloha_insertion.yaml \
  --save-video example_videos