Task-Aware Verifiable RNN-Based Policies for Partially Observable Markov Decision Processes.
Steven CarrNils JansenUfuk TopcuPublished in: J. Artif. Intell. Res. (2021)
Keyphrases
- partially observable markov decision processes
- optimal policy
- recurrent neural networks
- finite state
- reinforcement learning
- dynamical systems
- decision problems
- belief state
- continuous state
- markov decision processes
- planning under uncertainty
- dynamic programming
- sufficient conditions
- partially observable stochastic games
- belief space
- multi agent
- state space
- planning problems
- policy search
- partially observable domains
- partially observable
- bayesian reinforcement learning
- predictive state representations
- approximate solutions
- sequential decision making problems
- partially observable markov
- average reward
- decision processes
- infinite horizon
- belief revision
- np hard
- policy gradient
- dec pomdps
- dynamic environments