Semantic and geometric SLAM, visual-inertial state estimation, and IMU/GNSS fusion — built in C++ and evaluated on benchmark and self-collected datasets.

Structure from motion, pose-graph optimization with GTSAM, and camera calibration — the geometry underneath the SLAM work above.

Vision-language-action models and reinforcement learning applied to manipulation and control, with an emphasis on what breaks between a trained policy and a running robot.

Optimal control, classical computer vision, and deep learning coursework.