DexHand Lab Pro: Tactile Microsuture Dexterity Benchmark

[ducthuykh1009]

Registration UUID: 2555924c-74a4-4788-be61-1f1e65bf3f44

Project name: DexHand Lab Pro
Submission folder: submissions/dexhand_lab

Summary:
This update keeps DexHand Lab as a hand-only MuJoCo dexterous manipulation benchmark and adds a visible high-dexterity task: tactile microsuture threading. The five-finger hand performs an index probe, thumb-index-middle tripod pinch with ring stabilizer, two controlled needle passes through a tissue-pad target, thread-loop pull, and tension-limited no-tear verification. The task is shown in the generated demo video and logged in machine-readable evidence files.

What changed:

• Added microsuture needle, visible thread loop, tissue pad, entry/exit eyelet markers, and close-up suture camera to the MJCF scene.
• Added MICROSUTURE_THREADING grasp primitive with finger roles, closure sequence, stability conditions, and recovery strategy.
• Added microsuture object classification and affordance metadata.
• Added SUTURE_* phases to the main demo, HUD, narration, contact timeline, trajectory, summary, final report, and judge replay index.
• Added microsuture_benchmark.py plus dataset/microsuture_threading_report.json, dataset/microsuture_threading_trace.csv, and outputs/microsuture_scorecard.json.
• Expanded the deterministic verification suite from 39 gates to 45 gates.
• Tightened contact-causality/no-snap audit so generated evidence reports zero pre-verification motion events.
• Updated README, JUDGE_BRIEF, rubric_scorecard.json, submission_manifest.json, validator, and tests.

Validation performed:

• python submissions/dexhand_lab/run_demo.py
• python submissions/dexhand_lab/run_demo.py --episodes 3 --seed 42 --no-video --difficulty medium
• python submissions/dexhand_lab/run_stress_eval.py --seeds 32
• python submissions/dexhand_lab/arena_task_suite.py
• python submissions/dexhand_lab/microsuture_benchmark.py
• python -m unittest discover -s submissions/dexhand_lab/tests -p "test_*.py"
• python submissions/dexhand_lab/validate_submission.py

Final generated metrics:

• Overall task success: true
• Task gates: 45/45
• Demo duration: 85.25 s
• Object snap events: 0
• Contact-causality pass: true
• Verified motion frame rate: 1.0
• Average active fingers for dexterous grasps: 4.18272
• Average multi-side contact score for dexterous grasps: 0.92919
• Cap rotation: 224/224 deg
• Load hold: 9.0x
• Blind tactile classifier accuracy: 1.0
• Precision assembly success: true
• Combination lock success: true
• Microsuture threading success: true
• Microsuture passes: 2/2
• Microsuture entry/exit error: 0.0024/0.0028 m
• Microsuture tension: 0.42/0.65 N
• Local readiness estimate: 100/100

Honest scope:
The project uses MuJoCo simulation-native state/contact information, fingertip touch sensors, and deterministic/contact-aware heuristic control. It does not claim learned RL, real camera perception, physical tactile hardware, perfect contact physics, physical surgery readiness, or real-world deployment.

[ducthuykh1009]

Update pushed: aligned DexHand Lab evidence with event rules and fixed the readiness mismatch.

Changes in this update:

• Fixed summary/report ordering so outputs/summary.json, submission_readiness_report.json, and validator_report.json agree.
• Aligned the generated narration SRT and keyframe panel to the actual 85.25s demo video inside the required 1-3 minute window.
• Updated keyframe selection to use real trajectory phases, making microsuture, vial, lock, assembly, tactile, and hardware-audit evidence visible in the correct segments.
• Restored nested stress_eval_summary and dataset_stress_eval_path in summary.json while preserving top-level stress metrics.
• Updated README, Judge Brief, rubric scorecard, and quality-gate presentation text to match the current generated demo.

Re-tested locally:

• python submissions/dexhand_lab/run_demo.py
• python submissions/dexhand_lab/run_demo.py --episodes 3 --seed 42 --no-video --difficulty medium
• python submissions/dexhand_lab/run_stress_eval.py --seeds 32
• python submissions/dexhand_lab/arena_task_suite.py
• python submissions/dexhand_lab/microsuture_benchmark.py
• python -m unittest discover -s submissions/dexhand_lab/tests -p "test_*.py"
• python submissions/dexhand_lab/validate_submission.py

Current evidence highlights:

• validator_passed=true
• submission_readiness_pass=true
• rules_alignment_pass=true
• task gates: 45/45
• object_snap_events=0
• contact_causality_pass=true
• cap_rotation_success=true
• load_hold_success=true
• microsuture_threading_success=true