
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
Goal Condition DRL for Robotic Control
The next frontier that we need to address is to have systems that can handle a wide range of tasks. The thing that we settled on to start with was goal condition reinforcement learning, where essentially the robot gets in the early days, literally a picture of what the environment should be and it tries to manipulate the environment until it matches that picture. So you can define like a very simple thing like are you holding an object and lots of complexity merges from that just through kind of autonomous interaction.
Transcript
Play full episode