Abstract
Robotics is a fantastic testbed for both RL theory and RL practice. We have some problems (e.g. a humanoid doing a backflip) that look very hard, but model-based control performs quite well. We have other problems that really shouldn’t be that hard (e.g. spreading peanut butter on toast), where model-based control falls short. I’ll discuss a key theoretical sticking points; progress here might unlock practical success. I’m excited to see what you come up with this fall!