Reinforcement Learning of Risk-Constrained Policies in Markov Decision Processes.
Tomás BrázdilKrishnendu ChatterjeePetr NovotnýJiri VahalaPublished in: AAAI (2020)
Keyphrases
- markov decision processes
- optimal policy
- reinforcement learning
- risk sensitive
- markov decision process
- reward function
- reinforcement learning algorithms
- state space
- policy iteration
- policy iteration algorithm
- control policies
- total reward
- average cost
- finite state
- decision processes
- markov decision problems
- discounted reward
- decentralized control
- dynamic programming
- decision problems
- action space
- finite horizon
- long run
- state and action spaces
- average reward
- transition matrices
- expected reward
- multistage
- partially observable
- infinite horizon
- partially observable markov decision processes
- macro actions
- factored mdps
- decision making
- state abstraction
- stationary policies
- action sets
- function approximation
- reachability analysis
- stochastic games
- planning under uncertainty
- hierarchical reinforcement learning
- sufficient conditions
- initial state
- decision theoretic planning
- model based reinforcement learning
- continuous state
- markov games
- policy evaluation
- multi agent
- model free
- optimal control
- real time dynamic programming