Selecting the Partial State Abstractions of MDPs: A Metareasoning Approach with Deep Reinforcement Learning.
Samer B. NashedJustin SvegliatoAbhinav BhatiaStuart RussellShlomo ZilbersteinPublished in: IROS (2022)
Keyphrases
- reinforcement learning
- state space
- markov decision processes
- action space
- state abstraction
- markov decision process
- optimal policy
- transition model
- function approximation
- initial state
- machine learning
- hidden state
- partially observable
- temporal difference
- learning algorithm
- markov decision problems
- state transitions
- state and action spaces
- policy iteration
- infinite horizon
- state transition
- decision theoretic
- control problems
- reinforcement learning algorithms
- reward function
- temporal abstractions
- heuristic search
- factored mdps
- high level
- action sets