HyAR: Addressing Discrete-Continuous Action Reinforcement Learning via Hybrid Action Representation.
Boyan LiHongyao TangYan ZhengJianye HaoPengyi LiZhen WangZhaopeng MengLi WangPublished in: ICLR (2022)
Keyphrases
- continuous action
- policy search
- continuous state
- reinforcement learning
- action space
- state space
- partially observable markov decision processes
- continuous state and action spaces
- robot navigation
- function approximation
- reinforcement learning algorithms
- dynamic programming
- markov decision processes
- finite state
- reward function
- state dependent
- temporal difference
- real valued
- linear programming
- machine learning
- neural network
- control policies
- markov decision problems
- hidden state