vision robustnessvla-robustness-kit
Camera shift causes VLA failure cluster
A synthetic episode set shows brittleness when camera pose metadata changes between otherwise similar tabletop tasks.
Review label
vision-robustness
Expected finding
camera_shift appears as a repeated failure cluster
camera pose driftparaphrase brittlecluster representative
vla-robustness-kit run examples/mock_episode_set --policy mock --out report.mddataset QAlerobot-quality-gates
Low-quality teleop episode blocks training readiness
A synthetic SO-101 review case groups missing frames, timestamp drift, and action-state mismatch into one review decision.
Expected finding
episode is excluded until dropped frames and action gaps are reviewed
dropped framestimestamp driftaction gap
lerobot-quality-gates scan examples/so101_quality_gate --format markdown --out report.mdsensor QArobotics-reviewkit
Sensor desync masks repeated grasp failures
A public, synthetic case for reviewing RGB, joint-state, and language stream alignment before export.
Review label
sensor-desync
Expected finding
AV sync warning is attached before the reviewed subset is exported
AV syncjoint continuityexport blocker
robostudio inspect examples/mock_multi_format/lerobot_v3 --sensor-qa --out qa.mdintervention reviewrobot-recovery-bench
Recovery segment missing intervention label
A recovery benchmark case that keeps unlabeled intervention spans visible instead of hiding them in aggregate metrics.
Review label
recovery-metadata
Expected finding
manual recovery span must be labeled before benchmark scoring
manual recoveryphase labelbenchmark score
robot-recovery-bench score examples/recovery_episode --taxonomy intervention --out recovery.md