Simple Regret Optimization in Online Planning for Markov Decision Processes.
Zohar FeldmanCarmel DomshlakPublished in: J. Artif. Intell. Res. (2014)
Keyphrases
- markov decision processes
- planning under uncertainty
- online learning
- decision theoretic planning
- state space
- macro actions
- finite state
- dynamic programming
- optimal policy
- reward function
- partially observable
- transition matrices
- reinforcement learning
- factored mdps
- reachability analysis
- policy iteration
- model based reinforcement learning
- reinforcement learning algorithms
- partially observable markov decision processes
- decision processes
- finite horizon
- action space
- average reward
- expected reward
- total reward
- probabilistic planning
- infinite horizon
- lower bound
- markov decision process
- average cost
- ai planning
- planning problems
- dynamical systems
- markov decision problems
- state abstraction
- action selection
- domain independent