Description
RL agents usually need millions of real trials to learn games. DreamerV2 learns a discrete world model of the environment and trains its policy in imagination, greatly improving efficiency.
It was the first world-model agent to reach human-level performance on the Atari benchmark.
World model:Learned environment.
Imagination:Efficient learning.
Human-level:On Atari.
It was the first world-model agent to reach human-level performance on the Atari benchmark.
Features
World model:Learned environment.
Imagination:Efficient learning.
Human-level:On Atari.
