Minimal

I used Minimal as a fast, lightweight platform to validate the core components of PACE — especially actuator modeling, parameter identification, and sim-to-real locomotion transfer. Its small size, 3D-printed structure, and pseudo–direct-drive variable-gear actuators made it ideal for rapid experimentation, allowing me to iterate on models, rewards, and control strategies far more quickly than on larger systems.

On the real robot, I deployed the learned controllers to test robustness under challenging conditions relative to its scale: steep obstacles, uneven surfaces, and full stair climbs that match nearly the robot’s own height. Through repeated hardware experiments, I refined both the controller and the simulation fidelity, ultimately demonstrating that even a small, low-cost platform can achieve reliable stair ascent and agile terrain traversal when powered by accurate actuator models and a well-designed sim-to-real pipeline.

Beyond the research contributions, I also acted as Minimal’s “parent,” taking responsibility for its day-to-day reliability as a hardware platform. This included debugging actuator electronics, fixing mechanical failures in the 3D-printed structure, tuning low-level controllers, and ensuring that experiments could run consistently without unexpected hardware issues. By keeping the robot stable, calibrated, and operational, I enabled fast iteration cycles for sim-to-real experiments and made Minimal a dependable small-scale testbed for PACE.