Discrete Compositional Representations as an Abstraction for Goal Conditioned Reinforcement Learning.
Riashat IslamHongyu ZangAnirudh GoyalAlex M. LambKenji KawaguchiXin LiRomain LarocheYoshua BengioRemi Tachet des CombesPublished in: NeurIPS (2022)
Keyphrases
- reinforcement learning
- state abstraction
- model free
- optimal policy
- continuous state and action spaces
- learning process
- multi agent
- high level
- database
- robotic control
- object recognition
- reinforcement learning methods
- reinforcement learning algorithms
- decision theoretic planning
- learning algorithm
- continuous state spaces
- continuous state
- temporal abstractions
- control problems
- markov decision processes
- higher level
- least squares
- state space
- dynamic programming
- evolutionary algorithm