Episode 28: Sergey Levine, UC Berkeley, on the bottlenecks to generalization in reinforcement learning, why simulation is doomed to succeed, and how to pick good research problems
Generally Intelligent
00:00
Using Convolutional Network Techniques to Control Robot Motion
John Schulman: I started working on how we could kind of more or less throw that out and replace it with end-to-end learning from deep nets. He says his work resulted in a paper that was basically the first deep reinforcement learning paper for image-based real world robotic manipulation. "I intentionally want to make it like a little bit extreme," he adds.
Transcript
Play full episode
