name: simulation-data
description: "Routes simulated ALOHA episode generation, replay, visualization,
mirroring, compression, and truncation workflows for ACT++ HDF5 data."
disable-model-invocation: true
metadata:
disco-role: operating
license: MIT
simulation-data
Use this sub-skill when the task is about creating, replaying, visualizing, mirroring, compressing, truncating, or validating simulated ACT++ episode data.
Typical triggers
- "Generate 50 scripted transfer-cube episodes"
- "Replay episode_0.hdf5"
- "Mirror and compress the insertion dataset"
- "Why does sim reset assert on BOX_POSE?"
- "What fields are in the episode HDF5 file?"
What this sub-skill covers
- Joint-space and end-effector MuJoCo/DM Control environments.
- Scripted transfer-cube and insertion demo generation.
- Replaying HDF5 action traces back into simulation.
- Rendering videos and qpos plots from HDF5 episodes.
- Mirroring, JPEG compression, and truncation of episode files.
- Dataset layout and camera-name conventions for sim files.
What it excludes
- Model training and checkpoint management, which belong in policy-training.
- VINN feature caching and k-selection, which belong in vinn-offline.
- Real-robot alignment, servo diagnostics, and Mobile ALOHA deployment.
Read these first
Run this helper first
Before a long rollout, use check_sim_backend.py to confirm the sim backend can create and reset both environment variants from a chosen checkout.