Focuses on learning from demonstration and RL: causal inference to extract rewards/specifications, policy composition for long-horizon tasks, and integrating motion planning with neural controllers. Also works on video-based dynamics estimation and medical robotics applications.