A model based approach to exploration of continuous-state MDPs using Divergence-to-Go.
Matthew EmighEvan KrimingerJosé Carlos PríncipePublished in: MLSP (2015)
Keyphrases
- continuous state
- reinforcement learning
- continuous state and action spaces
- action space
- policy search
- markov decision processes
- finite state
- continuous state spaces
- action selection
- partially observable markov decision processes
- robot navigation
- state space
- control policies
- optimal policy
- state dependent
- planning problems
- real valued
- continuous action
- dynamic programming
- markov chain
- reinforcement learning algorithms
- markov decision process
- function approximation
- finite horizon
- policy iteration
- state action
- machine learning
- learning algorithm
- control policy
- single agent
- dynamical systems