Safe Reinforcement Learning through Meta-learned Instincts.
Djordje GrbicSebastian RisiPublished in: CoRR (2020)
Keyphrases
- reinforcement learning
- function approximation
- reinforcement learning algorithms
- model free
- learned knowledge
- temporal difference
- state space
- robotic control
- stochastic approximation
- evolutionary algorithm
- optimal policy
- optimal control
- learning algorithm
- temporal difference learning
- reinforcement learning methods
- meta level
- dynamic programming
- domain knowledge
- partially observable
- markov decision process
- genetic algorithm
- previously learned
- autonomous learning
- policy search
- real time