"Use high-level dm_control locomotion, soccer, mocap, walker, prop,
Scanned 9/8/2026
Install to Claude Code
npx -y skills add VectorSpaceLab/AREX-Skill --skill locomotion-manipulation --agent claude-codeInstalls into .claude/skills of the current project.
Are you the author of Locomotion Manipulation?
Add the live security badge to your README — it updates automatically with every re-scan.
[](https://www.skillsdirectory.com/skills/vectorspacelab-locomotion-manipulation)More formats (shields.io, HTML) on the badges page.
---
name: locomotion-manipulation
description: "Use high-level dm_control locomotion, soccer, mocap, walker, prop,
and manipulation task families."
metadata:
disco-role: operating
disable-model-invocation: true
license: Apache 2.0
---
# dm_control locomotion and manipulation router
Use this sub-skill when the task is about selecting or using built-in high-level `dm_control` manipulation, locomotion, soccer, walker, mocap, arena, or prop families. These are Composer-backed environments and components; they can be stepped with the same `dm_env` loop style as suite tasks once an environment is constructed.
## Route first
- For ready-made manipulation registry tasks, read [references/manipulation-reference.md](references/manipulation-reference.md) to choose `features` versus `vision`, validate names/tags, load with `dm_control.manipulation.load`, and smoke-test specs.
- For locomotion examples, walkers, arenas, soccer, mocap/reference-pose tracking, and asset/data caveats, read [references/locomotion-reference.md](references/locomotion-reference.md) before constructing or importing heavier task families.
- For common failures around unknown names/tags, feature/vision confusion, missing HDF5/labmaze/assets, expensive demos, and rendering requirements, read [references/troubleshooting.md](references/troubleshooting.md).
- Run [scripts/list_locomotion_manipulation.py](scripts/list_locomotion_manipulation.py) to list installed manipulation tasks/tags and verify that common locomotion example modules import; add `--smoke-manipulation` for a one-reset/one-step manipulation smoke.
## Boundaries
- Generic Control Suite tasks and benchmark loops belong in `../suite-rl-workflows/SKILL.md`.
- Building new Composer tasks/entities, custom observables, or reusable robot abstractions from scratch belongs in `../composer-environments/SKILL.md`.
- Rendering backend operations, viewer windows, and pixel-render troubleshooting belong in `../rendering-viewer-assets/SKILL.md`.
## Quick decision checklist
1. If the user names a manipulation task or asks for Jaco arm/hand tasks, validate it against `dm_control.manipulation.ALL` and load it with `dm_control.manipulation.load(name, seed=...)`.
2. If the user wants low-dimensional observations, prefer manipulation names ending in `_features`; if they want camera observations, use `_vision` and verify rendering support separately.
3. If the user names CMU humanoid, rodent, soccer, maze, corridor, bowl, target, or mocap/reference-pose tracking, treat it as locomotion/Composer-backed and inspect specs with a short reset/step before any long rollout.
4. If the task requires new arenas, reward functions, entities, or observables beyond choosing/configuring built-ins, route to Composer rather than extending this catalog inline.
Is this your skill, or is something wrong with this listing? Request removal or report an issue. Author removals are honored within 72 hours.
No comments yet. Be the first to comment!