Danijar Hafner 2
TalkRL: The Reinforcement Learning Podcast
Dreamer V2: A Novel Algorithm for Continuous Control
In dreamer v1 we focused on continuous control from pixels. And it was very data efficient and got high final performance. But also pretty narrow in the sense that we couldn't deal with the street actions very well. We weren't competitive on standard benchmarks like Atari. My goal for dreamer v3 was to automate all of that away. So you can have an algorithm where the objective functions are robust enough that you can just run it out of the box and it'll give you good performance.
00:00
Transcript
Play full episode
Remember Everything You Learn from Podcasts
Save insights instantly, chat with episodes, and build lasting knowledge - all powered by AI.
