@Motion planning and controls, Robot autonomy team

@Motion planning and controls, Robot autonomy team

• Demonstrated autonomy software features to modulate vehicle speed for rapid changes and discontinuities in off-road terrain • Built negative obstacle regression simulation set to validate planning behavior

• Integrated human behavior prediction models with reachability to maintain optimally conservative safety monitor (Backward Reachable Tube for active collision-avoidance) in interactive autonomous driving

Lunar CubeRover - NASA PFP

• Formulated optimal control strategies under uncertainties for effective docking and achieved 100% success rate • Incorporated external uncertainty learning with expert demos making EKF quickly adapt to extraterrestrial environments

Level 4 autonomy @Robot Learning Lab, University of Washington

• Executed Cross-entropy and Inverse RL to auto-tune the planner’s cost function saving 5 hours of manual tuning • Developed feature debugger that quantitatively compares nominal planner decisions with expert demonstrations • Implemented parallel version for cross-track error in MPC-based local planning that is 115X faster than serial computation