Tytan

I used Tytan as the primary platform to demonstrate large-scale sim-to-real transfer, deploying reinforcement-learning locomotion controllers trained entirely in simulation onto a real, torque-dense quadruped. My work focused on building the full simulation pipeline, developing high-fidelity actuator models, and validating that learned policies generalize to high-speed locomotion, heavy payloads, and diverse terrain conditions on the real robot.

At the same time, I acted as Tytan’s “parent”: I was responsible for the robot’s overall reliability — from software integration and low-level controller debugging to diagnosing electrical issues, sensor problems, and mechanical failures. Much of the project involved making Tytan a robust, everyday-usable research platform, ensuring that experiments ran smoothly and that the learned controllers could be evaluated safely and consistently in the real world.