Fitted Q-iteration in continuous action-space MDPs.
András AntosRémi MunosCsaba SzepesváriPublished in: NIPS (2007)
Keyphrases
- action space
- fitted q iteration
- reinforcement learning
- markov decision processes
- state space
- state and action spaces
- continuous state spaces
- continuous state
- real valued
- markov decision process
- control policies
- state action
- markov decision problems
- stochastic processes
- optimal policy
- finite state
- continuous action
- learning algorithm
- control problems
- action selection
- reinforcement learning algorithms
- single agent
- computational complexity
- function approximation
- partially observable
- policy iteration
- heuristic search
- function approximators
- average reward
- temporal difference
- planning under uncertainty
- optimal control
- probabilistic model
- decision making
- machine learning