"Export ProtoMotions trackers, validate deployment input contracts,
Scanned 9/8/2026
Install to Claude Code
npx -y skills add VectorSpaceLab/AREX-Skill --skill deployment-and-robots --agent claude-codeInstalls into .claude/skills of the current project.
Are you the author of Deployment And Robots?
Add the live security badge to your README — it updates automatically with every re-scan.
[](https://www.skillsdirectory.com/skills/vectorspacelab-deployment-and-robots)More formats (shields.io, HTML) on the badges page.
---
name: deployment-and-robots
description: "Export ProtoMotions trackers, validate deployment input contracts,
test MuJoCo deployment, and add or troubleshoot robot assets."
disable-model-invocation: true
metadata:
disco-role: operating
license: Apache 2.0
---
# ProtoMotions deployment and robots
Use this sub-skill for G1 deployment tracker workflows, ONNX export, standalone MuJoCo validation, tracker input semantics, real-robot integration cautions, custom robot configs, MJCF/USD conversion, and cross-simulator robot issues.
## Read first
- `references/deployment-contract.md`: ONNX tracker inputs, frame conventions, MuJoCo validation, and real-robot safety.
- `references/onnx-export.md`: export artifact roles and checkpoint/config requirements.
- `references/custom-robots-and-assets.md`: adding robots, asset requirements, MJCF/USD, and factory registration.
- `references/troubleshooting.md`: deployment, input-frame, robot asset, and simulator migration failures.
- `scripts/inspect_tracker_yaml.py`: summarize a deployment YAML sidecar.
- `scripts/tracker_alignment_smoke.py`: pure NumPy smoke for the heading/reference alignment contract.
## Deployment action pattern
1. Verify the checkpoint directory includes `resolved_configs_inference.pt` or an accepted fallback config.
2. Export the policy to `unified_pipeline.onnx` plus `unified_pipeline.yaml` using a CPU-capable environment with PyTorch and ProtoMotions; no simulator should be needed for export.
3. Inspect YAML input/output metadata before wiring a deployment framework.
4. Validate in standalone MuJoCo before real hardware.
5. Preserve heading alignment, root-local angular velocity frame, action post-processing, acceleration clamp, EMA action filter, and PD target semantics.
6. Require explicit human approval and safety procedures for real robot execution.
## Custom robot action pattern
1. Provide MJCF as the source-of-truth robot asset.
2. Define robot config: body-name mappings, trackable body subset, asset config, default root height, PD controls, and per-simulator params.
3. Register the robot name in the factory.
4. Run factory/config smokes before simulator instantiation.
5. Validate in a random-pose or simple simulation check before training.
6. For IsaacLab, handle MJCF-to-USD conversion and D6 workaround behavior carefully.
## Non-negotiable safety
Real-robot deployment can cause unsafe physical motion. Never skip MuJoCo validation, PD/action post-processing checks, emergency-stop verification, or human authorization when moving from simulation to hardware.

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!