Provably Safe Model-Based Meta Reinforcement Learning: An Abstraction-Based Approach.
Xiaowu SunWael FatnassiUlices Santa CruzYasser ShoukryPublished in: CoRR (2021)
Keyphrases
- reinforcement learning
- model free
- state abstraction
- high level
- function approximation
- state space
- reinforcement learning algorithms
- data driven
- temporal abstractions
- temporal difference
- meta level
- robotic control
- theoretical guarantees
- supervised learning
- data mining
- neural network
- optimal policy
- optimal control
- mobile robot
- clustering algorithm
- learning algorithm
- machine learning
- fully unsupervised
- data abstraction
- stochastic approximation
- meta reasoning
- data sets