Add maximum-height jump policy

[ThomasBurgess2000]

What this PR does

Adds a distinct Maximum-Height Jump community behavior without replacing the existing Vertical Jump entry.

Preview

preview-4x.mp4

20 seconds, 4× slow motion: standing start, maximum-height jump, landing, and standing-policy handoff.

• Publishes the 61D-input, 14-action, 50 Hz baked-normalizer ONNX from a commit-pinned public artifact repository.
• Includes a verified 20-second, 4× slow-motion preview rendered from the exact published ONNX pair.
• Labels the policy community_experimental and simulation-only; no hardware claim is made.
• Documents that the canonical jump ONNX lands but does not return to a durable standing pose by itself.
• Documents the optional two-policy reset: a smoothstep blend to the companion standing ONNX beginning at 0.22 s over 0.14 s.

Simulation evidence

Direct ONNX Runtime evaluation in one seeded CPU MuJoCo rollout measured:

• 0.628 m/s launch velocity
• 31.67 mm whole-body visible rise
• 31.67 mm bilateral sole clearance
• 140 ms airtime

The jump controller alone had stable_landing=true and durable_landing=false. The documented handoff had stable_landing=true, durable_landing=true, failed_recovery=false, no non-foot body contact, and a 100 ms recovery margin.

Artifact and reproducibility evidence: ThomasBurgess2000/microduck-max-height-jump

Checklist

• pnpm validate passes
• pnpm test passes
• pnpm compile was run and public/registry.json is committed
• The source, license, compatibility, and verification label are accurate
• Community work is not presented as official Pollen Robotics work
• pnpm build passes

[chatgpt-codex-connector]

You have reached your Codex usage limits for security reviews. Please try again later.

[chatgpt-codex-connector]

Codex Review Summary

This comment shows the latest Codex review activity on this pull request.

Review	Status	Commit	Review trigger
📝 Code Review	✅ Completed 2026-08-31T17:32:37.870351Z	62d1bd8	PR opened

ℹ️ About Codex in GitHub

Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you

• Open a pull request for review
• Mark a draft as ready
• Comment "@codex review" or "@codex security review".

Codex reacts with 👀 while any review is running, comments if it has suggestions, and reacts with 👍 once all reviews finish with no findings.

[ob1-s]

hey @ThomasBurgess2000, thank you so much for adding this move to the registry - it’s our first external contribution 🎉 it genuinely means a lot to me.

if you (or your AI agent haha) have a minute, i'd love to hear how the contribution experience went: was anything confusing or annoying, did you get stuck anywhere, or was there anything you expected this registry to do that it didn't?

also curious what made you decide to add your policy here in the first place. no need for a detailed answer - any feedback would be super useful :)

[ThomasBurgess2000]

hey @ob1-s ! I just saw your message in discord about this project and told 5.6 Sol to try and add the policy I had trained to it for fun :) I asked it if there were any notes for the maintainer re: things that would have made the contribution easier:

1. Fix and document pnpm cli submit.
   The command exists but is not mentioned in CONTRIBUTING.md. It commits only the behavior descriptor and says CI will rebuild the catalog, but CI fails when the regenerated public/registry.json is not committed. It should produce a complete, CI-passing PR or clearly be labeled descriptor-only.

2. Add a descriptor scaffold and aggregate preflight.
   A command such as pnpm cli init <id> could generate the fixed 61D/14-action contract and prompt for licensing, verification, artifacts, command semantics, and media. A second command should run validation, tests, compilation, and build together.

3. Make artifact validation match the CI job’s name.
   “Validate Registry & Artifacts” currently validates primarily the JSON schema. Useful additional checks would include:

   • Artifact URL accessibility and content type
   • Optional SHA-256 checksums
   • ONNX input/output dimensions (61 → 14)
   • Local media-file existence
   • Video codec and decode checks

4. Document and automate the media contract.
   CONTRIBUTING.md does not explain loop_url versus video_url, recommended duration, codec, aspect ratio, file location, or size limits. PR Add maximum-height jump policy #11 initially used one 20-second slow-motion video for both; a separate short card loop was added later. A media-preparation command could generate both automatically.

5. Generate the README catalog.
   The README’s behavior count and list are a second manually maintained source of truth. They should be generated from the descriptors alongside public/registry.json.

6. Unify the PR template and generated PR body.
   The CLI and .github/PULL_REQUEST_TEMPLATE.md currently use different checklists. A shared template could include sections for preview media, artifact provenance, verification evidence, command semantics, and known limitations.

7. Define a standard release manifest.
   A small release.json format could describe the canonical ONNX, checksum, license, source revision, metrics, media, and optional companion policies. The registry CLI could consume it to generate most of the descriptor automatically.

8. Document first-time fork CI approval.
   The workflow correctly waited for maintainer approval, but contributors should be told to expect the action_required state.

[ob1-s]

hey, just wanted to thank you again for the feedback :)

I just merged an onboarding overhaul - turned out the cli bloat was most of the friction, so it's gone for now. agents can just do their stuff

also, your policy has its own page now: https://uduckmoves.com/behaviors/max-height-jump

(every policy/behavior link gets its own Open Graph card with a randomized duck pose 🦆)