Reinforcement Learning When All Actions Are Not Always Available.
Yash ChandakGeorgios TheocharousBlossom MetevierPhilip S. ThomasPublished in: AAAI (2020)
Keyphrases
- reinforcement learning
- action selection
- perceptual aliasing
- state and action spaces
- partially observable
- action space
- partially observable domains
- markov decision processes
- function approximation
- state space
- model free
- state action
- multiagent reinforcement learning
- partial observability
- learning algorithm
- action sets
- decision theoretic
- dynamic programming
- macro actions
- initially unknown
- agent receives
- behavioural cloning
- learning agent
- learning capabilities
- reward function
- goal directed
- plan recognition
- optimal policy
- multi agent systems
- reasoning about actions
- temporal difference learning
- action models
- sensory inputs
- continuous state
- reinforcement learning algorithms
- human actions
- robotic control
- multi agent