Deep Recurrent Q-Learning for Partially Observable MDPs.
Matthew J. HausknechtPeter StonePublished in: AAAI Fall Symposia (2015)
Keyphrases
- partially observable
- reinforcement learning
- state space
- markov decision processes
- reward function
- discount factor
- policy iteration
- markov decision problems
- reinforcement learning algorithms
- optimal policy
- partial observability
- stochastic shortest path
- function approximation
- continuous state spaces
- dynamic programming
- decision problems
- infinite horizon
- partial observations
- markov decision process
- heuristic search
- partially observable environments
- dynamical systems
- action models
- markov chain
- finite state
- state variables
- probabilistic planning
- continuous state and action spaces
- model free
- belief state
- orders of magnitude
- multi agent
- learning algorithm
- decision processes
- action selection
- planning under uncertainty
- partially observable markov decision process
- policy search
- inverse reinforcement learning
- search space
- special case
- lower bound
- action space
- search algorithm
- continuous state
- learning agent
- average cost
- transition probabilities