San Francisco Bay Area
[Python, NVIDIA Isaac Lab, ROS2, PyTorch] • Developed sim-to-real reinforcement learning policies for humanoid loco-manipulation, utilizing a custom hierarchical framework that integrates dexterous manipulation with whole-body control.
[Python, NVIDIA Isaac Gym, ROS2, PyTorch] • Developed DextrAH-G: a safe, continuously reacting pixels-to-action policy that achieves state-of-the-art dexterous grasping in the real world and was trained entirely in simulation using RL and teacher-student distillation with a geometric fabric controller
[Python, C++, NVIDIA Omni Isaac Gym] • Developed an obstacle-aware global navigation system for robotic arms using RMPflow and learning • Used a combination of deep reinforcement learning and supervised learning to move a robot arm through complex obstacle distributions without getting stuck in local minima
[Python, ROS, Simulink, MATLAB, Bash, Git] • Led the control software team to develop a pathfinding system for the world’s first fully autonomous sailboat that will sail on the 4,200 km journey from Victoria to Maui, with a sub-team of six controls team members in a full team of forty students • Spearheaded an RRT* pathfinding system that generates paths over 500% faster than previous systems,which improves the boat’s overall speed, responsiveness, and pathfinding quality • Implemented and tuned a controller using Model Predictive Control, which has the ability to anticipate future events and can take control actions accordingly, making it robust to significant changes to environmental conditions • Created a mathematical model of our sailboat's sailing and wind dynamics using Simulink and MATLAB • Performed extensive automated testing of our control system using a Gazebo sailboat simulator • Integrated the path-planning modules, sensors, and actuators with the control system using ROS
[C++, Python, Bash, Git] • Developed safety-critical C++ software for Tesla’s Full Self-Driving Beta system for planning and trajectory optimization