Multi-modal sensor fusion under dynamic occlusion
Combining vision, LiDAR, and tactile feedback to maintain object state when the gripper or operator blocks the camera.
ResearchResearch that
Research that
ships to the floor
Our research agenda is shaped by the failure modes we observe in live deployments. Every paper we publish corresponds to a problem one of our partners is solving in production right now.
Force-aware imitation learning from operator demos
Distilling tactile + force feedback into action policies that handle deformable and fragile objects.
Sim2Real with curated real-world data anchors
Using small, high-fidelity real datasets to align large simulated training corpora and reduce reality gap.
Operator hand-off and shared autonomy in production
Mid-task hand-off protocols between humans and policies, with confidence-driven role switching.
Selected Publications
Recent work
- 2026ICRATactile-conditioned policies for unstructured manipulation
- 2025CoRLDistribution-aware data scaling for Physical AI
- 2025RSSOperator-in-the-loop imitation learning from active manufacturing facilities
- 2024NeurIPS WorkshopSynthetic + real curriculum for multi-finger grasping