Code as Policies: What Generated Robot Code Needs to Run
Code as Policies needs more than valid Python: check every generated helper, coordinate convention and feedback loop before connecting robot APIs.
Physical AI Lab
Touch100k: Choosing Tactile Language Labels for Your Task
Touch100k preserves short tactile-property phrases and richer contact descriptions. Match each label level to the query and evaluate them on separate, object-aware splits.
Sparsh: When to Freeze a Tactile Encoder
A frozen Sparsh encoder can reduce label demand for a new touch task, but compare it with end-to-end training on the same sensor, preprocessing and held-out episodes.
BEHAVIOR 2026: Separating Policy Tuning from Evaluation
BEHAVIOR 2026 assigns public indices 0–9 to reported rollouts and 10–19 to pre-final testing. Freeze the policy, evaluator and timeout before reporting a score.
ManiSkill 3: Comparing Visual Simulator Throughput Fairly
FPS matters only after the scene, cameras, image channels and omitted pipeline work match. Use a workload table to compare ManiSkill 3 visual throughput and memory.
SIMPLER: Diagnosing Sim-to-Real Policy Ranking Mismatches
When simulation and real trials rank robot policies differently, trace the first divergence through model inputs, action conversion and executed motion before selecting a checkpoint.
Octo on a new robot: bind the checkpoint inputs and actions
Map Octo’s cameras, proprioception, history masks, task conditioning and action head to a new robot before tuning or rollout.
LIBERO suites: choose the right transfer test
Spatial, Object, Goal and the 90-to-10 task boundary test different transfer claims. Preserve exposure, order and fixed starts before comparing policies.
UMI demonstration transfer: what handheld robot data must preserve
UMI-style handheld demonstrations need more than video. Check relative gripper trajectories, latency alignment, two-hand pose and robot feasibility before policy training.
FoAR force feedback: when contact should change a robot policy
FoAR gates force/torque information around predicted contact. Diagnose the first contact transition before adding an always-on force stream to a robot policy.