Stanford University
Emma Brunskill's course covers tabular RL, deep Q-learning, policy gradients, exploration, offline RL, and RLHF with public lecture slides and assignments. A strong theory-plus-practice sequel to David Silver's lectures.
Advanced learners with ML and probability. This is an advanced course in Reinforcement Learning & Robotics, running about 10 weeks (one quarter).
Free. Prior-year lecture videos (e.g., Winter 2019 and 2024) are public on Stanford Online's YouTube channel; slides and assignments are public.
Sergey Levine's graduate course goes deep on policy gradients, actor-critic, model-based RL, offline RL, exploration, and inverse RL with PyTorch homework. For learners aiming at RL research or robotics who want the most advanced free syllabus available.
Train agents from Q-learning to PPO and multi-agent setups in Colab, publish them to the Hub, and compete on leaderboards. Ideal for learners who want deep RL to be practical and fun from the first unit, with a free certificate at the end.
David Silver's classic lectures cover MDPs, dynamic programming, Monte Carlo and TD learning, function approximation, policy gradients, and planning, closely tracking the Sutton and Barto textbook. Still the standard first RL course for anyone serious about the field.