Portfolio

Indy Autonomous Challenge

On Caltech's Indy Autonomous Challenge team, I design, implement, and deploy learning-based planning & control algorithms onto our full-sized Indy racecar.

My research focuses on ensuring sample-efficient policy optimization for robot learning directly on hardware. New theoretical analysis guarantees both expected policy improvement and improved dynamics estimation on an iteration-to-iteration basis.

Check out the TV broadcast from our latest competition.

Space Debris Redirection

We command two controlled spacecraft to capture and redirect a third, uncooperative agent. In real time, with no pre-training, our algorithm discovers a solution that intercepts, captures, and redirects the uncooperative spacecraft by planning through the dynamics of the tether and the contact forces between the tether and uncooperative craft.

DARPA EVADE

In collaboration with NASA JPL and Method Aerospace, we developed and tested a custom tilt-jet VTOL aircraft powered by four JetCat P550 jet engines.

Our work developed a new thrust allocation algorithm to overcome the strong nonlinearities in the tilt mechanism and strong actuator windup effects in the jets.