Reinforcement learning is the corner of machine learning that most courses skip, and when people do reach for it this lecture series is still where a lot of them start, for good reason. David Silver was central to the AlphaGo work, and teaching this material he has the rare quality of making genuinely hard ideas feel inevitable rather than intimidating. He builds it in the order the subject actually wants, starting with Markov decision processes so you understand the problem being solved, then dynamic programming, then Monte Carlo and temporal difference methods for learning from experience, then function approximation once tables stop scaling, and finally policy gradient methods. By the end you have a real mental model of what an agent, a reward, a value function, and a policy are, and why the algorithms are shaped the way they are, which is exactly the grounding you need before anything modern makes sense.
It pairs naturally with the Sutton and Barto book, and the two together are close to a complete theoretical education in the field for no money at all. Be clear eyed about what it is not. The recordings are old, so the deep reinforcement learning era, the methods that pair RL with large neural networks and the techniques behind recent results, gets little to no coverage, and you will need a more current resource for that. It is also a theory course at heart, dense with notation and proofs of intuition, and it deliberately does not hand you code, so you should be implementing the algorithms yourself in parallel or following a practical course alongside it, otherwise the ideas stay abstract.
And like any lecture series there is no grading, no problem sets handed back, and no cohort, the structure is entirely on you. None of that dislodges it from the top spot for foundations. Watch this for the why, write code elsewhere for the how, and you will understand reinforcement learning far better than people who only ever ran someone else's training script.