Understanding Failures of Deterministic Actor-Critic with Continuous Action Spaces and Sparse Rewards.
Guillaume MatheronNicolas PerrinOlivier SigaudPublished in: ICANN (2) (2020)
Keyphrases
- action space
- reinforcement learning
- actor critic
- markov decision processes
- policy iteration
- state space
- reinforcement learning algorithms
- temporal difference
- average reward
- function approximation
- action selection
- state action
- continuous action
- policy gradient
- reinforcement learning methods
- dynamic programming
- machine learning
- model free
- optimal policy
- partially observable
- finite state
- real valued
- markov decision process
- markov decision problems
- multi agent
- average cost
- optimal control
- least squares
- planning problems
- stochastic processes
- objective function