Robotics

Reinforcement learning teaches a robot to succeed by trial and error against a reward signal you have to design. Imitation learning skips that design problem entirely: you show the robot what success looks like, and it learns to reproduce it. This course covers how demonstrations are collected, why the most direct approach (behavioral cloning) breaks down in a specific and predictable way, and the techniques, DAgger and inverse reinforcement learning, that were built to fix it. It closes with how imitation and reinforcement learning are combined in real systems rather than treated as competing choices.