Safe Policy Improvement in Constrained Markov Decision Processes.
Luigi BerducciRadu GrosuPublished in: CoRR (2022)
Keyphrases
- markov decision processes
- optimal policy
- policy iteration
- markov decision process
- average reward
- finite horizon
- average cost
- state space
- state and action spaces
- infinite horizon
- action space
- partially observable
- finite state
- decision processes
- dynamic programming
- reward function
- transition matrices
- markov decision problems
- reinforcement learning
- policy iteration algorithm
- control policies
- factored mdps
- decision theoretic planning
- partially observable markov decision processes
- expected reward
- decision problems
- total reward
- policy evaluation
- discounted reward
- model based reinforcement learning
- reachability analysis
- state dependent
- sufficient conditions
- continuous state spaces
- semi markov decision processes
- approximate dynamic programming
- planning under uncertainty
- model free
- reinforcement learning algorithms
- action sets
- multistage
- initial state
- stationary policies
- least squares
- search algorithm