Robotics

The capstone of the Perception & Autonomy track, this course introduces reinforcement learning as a fundamentally different way to produce robot behavior, one where a policy is learned from millions of trial-and-error attempts in simulation rather than derived by an engineer, as with the PID loops and motion planners covered earlier on this site. Built for learners who already understand localization, path planning, multi-robot coordination, and feedback control, it covers the state-action-reward-policy framing of an RL problem, why simulation and domain randomization are essential to training and transferring these policies, and when RL genuinely earns its cost over a hand-designed controller.