Posts

Manually tuning rewards to create human-like motion is an incredibly difficult task. I’ve spent the past few weeks going through some major works to better understand how you can use motion capture data to train policies that achieve human-like behavior. This blog covers only a small number of motion tracking papers. I hope to go over sim2real motion tracking efforts in a future blog post. In the meantime, I encourage you to do some searching on the latest works on training human-like motions using motion tracking! ...

A while back I wrote a blog on understanding the fundamentals of RL. I’ve spent the past couple weeks reading through Kevin Murphy’s Reinforcement Learning textbook and Sutton and Barto to review some of my fundamentals. This blog contains some notes to cover topics I haven’t yet talked about in my first attempt at explaining RL! What is Reinforcement Learning? Reinforcement Learning is all about the idea of interacting with your environment to learn good behaviors. Given the full state $s_t$, observation $o_t$, some policy $\pi$, action $a_t = \pi(o_t)$, and reward $r_t$, the goal of an agent is to maximize the sum of its expected rewards: ...