What makes useful auxiliary tasks in reinforcement learning: investigating the effect of the target policy.
Banafsheh RafieeJun JinJun LuoAdam WhitePublished in: CoRR (2022)
Keyphrases
- reinforcement learning
- optimal policy
- action selection
- previously learned
- policy search
- markov decision processes
- function approximation
- markov decision process
- approximate dynamic programming
- dynamic programming
- transfer learning
- temporal difference
- real robot
- policy evaluation
- transferring knowledge
- partially observable environments
- neural network
- reinforcement learning problems
- markov decision problems
- partially observable domains
- control policies
- multiple tasks
- function approximators
- action space
- partially observable
- reinforcement learning algorithms
- infinite horizon
- state space
- machine learning