fix(libero): expose grasp_status and mark pi0_pick success as motion heuristic only

[akushonkamen]

Description

Fixes #243.

Root cause. pi0_pick reported success from a pure kinematic heuristic with no contact or held-object evidence:

• robots/libero/tools.py:213 — default gripper_open_thresh=0.0 makes the closure lower bound ineffective;
• robots/libero/tools.py:256 — any grip < 0.06 counts as "closed";
• robots/libero/tools.py:257-259 — descent-done + post-descent ascent + closure ⇒ success=True.

So closing the fingers on air also returned success=true and misled downstream transport/release decisions.

Fix (per the issue's Expected section; no existing behavior changed).

• ToolResult data gains an honest grasp_status: "unknown" field (robots/libero/tools.py:273) — no contact/held-object query exists, and no privileged object state is exposed.
• Tool description and Args rewritten: success is motion-heuristic-only, a HINT not proof (robots/libero/tools.py:216).
• evaluate.py Rule 1b wording aligned to cite grasp_status (robots/libero/prompts/evaluate.py:209,219).
• tests/unit_tests/robots/fixtures/tool_schemas.json contract fixture updated (libero/pi0_pick: description + 3 arg descriptions — intentional declaration change).
• 7 new offline probe tests replaying the issue-243 probe table (z: 1.00→0.85→0.91, grip: 0.080→0.002→0.002) against stubbed VLA/env boundaries: empty grasp, closure lower-bound threshold, no-lift, episode-end paths + no-privileged-field contract + docstring/prompt-contract alignment.

Deliberately not done: simulation-based contact queries — the issue asks to align on that in the issue first, so it is not done unilaterally.

Testing

• Red: timeout 300 .venv/bin/python -m pytest tests/unit_tests/robots/libero/test_pi0_pick_grasp_status.py -q → 7 failed (all missing grasp_status / contract mismatch).
• Green: timeout 300 .venv/bin/python -m pytest tests/unit_tests/robots/libero/ -q → 52 passed (baseline 45 + 7 new).
• Full suite: main baseline timeout 600 pytest tests/unit_tests/ -q → 1209 passed, 7 skipped, 0 failed; this branch, same command → 1216 passed, 7 skipped, 0 failed (FAIL set empty both rounds; only +7 new tests, zero regressions).
• Lint: ruff check + ruff format --check (3 changed Python files) → All checks passed.
• Re-verified at push time on 064a179 (origin/main unchanged, no rebase needed): green suite 52 passed, new test file 7 passed, ruff check/ruff format --check on the 3 changed Python files → All checks passed.

Manual verification

Not needed — offline unit tests fully cover the change (no GPU, simulator, or model service involved; simulation-based contact verification is intentionally deferred pending discussion in #243).

Checklist

• My code follows the code style of this project.
• I have updated related documentation when needed.
• I have added tests to cover my changes when needed.
• All new and existing tests passed.

_{Developed with AI assistance; full disclosure in the claim comment on #243.}

[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	⚠️ Failed 2026-10-11T11:10:02.875311Z	064a179	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.