§ The Record
I build robots that learn from interaction — turning messy, real-world experience into behaviour reliable enough to trust, across robot learning, reinforcement learning, perception, and sim-to-real transfer. What follows is the complete record, newest first; filter it by focus, or narrow to the selected highlights.
Conditioning robot policies on camera pose via per-pixel ray embeddings makes them robust to viewpoint changes.
A flow-based module that corrects only the out-of-distribution actions of an existing robot policy, without retraining it.
A single-actuator wearable haptic device that renders directional touch cues to improve contact-rich robot teleoperation and imitation learning.
Distills diffusion-based shared autonomy into a one-step consistency model for real-time robot assistance.
No publications match these filters.