Reinforcement Learning

A four-cable parallel robot for air hockey. Four corner-mounted MJ5208 BLDC motors tension cables that meet at the mallet; an overhead camera feeds a mallet EKF and a puck predictor; a receding-horizon (naive MPC) planner picks between defending and quintic-spline strikes every 10 ms.

Moving ILT from slow physics-based gradient descent toward conditional latent diffusion and GRPO fine-tuning, with a dual-stage architecture for offline curvilinear mask generation and real-time EUV laser dose compensation.

ME 193B project exploring constrained RL for humanoid locomotion on dynamic stepping stones using IsaacLab and MuJoCo.