GPU-Accelerated Value Iteration for the Computation of Reachability Probabilities in MDPs.
Zhimin WuErnst Moritz HahnAkin GünayLijun ZhangYang LiuPublished in: ECAI (2016)
Keyphrases
- gpu accelerated
- markov decision processes
- state space
- markov decision process
- optimal policy
- dynamic programming
- reinforcement learning
- factored mdps
- policy iteration
- finite element
- stochastic shortest path
- real time
- finite state
- probability distribution
- partially observable
- partially observable markov decision processes
- heuristic search
- infinite horizon
- planning under uncertainty
- belief state
- average cost
- average reward
- markov decision problems
- markov chain
- least squares
- transition probabilities
- action space