Description
RL algorithms need retuned hyperparameters for every new task, generalizing poorly. DreamerV3 is a general reinforcement learning algorithm based on world models that masters diverse domains with one set of hyperparameters.
It was the first to collect diamonds in Minecraft from scratch without human data, published in Nature.
General:One hyperparameter set.
World model:Learns in imagination.
Minecraft:Diamonds from scratch.
JAX:Efficient.
It was the first to collect diamonds in Minecraft from scratch without human data, published in Nature.
Features
General:One hyperparameter set.
World model:Learns in imagination.
Minecraft:Diamonds from scratch.
JAX:Efficient.
