Correct Me If I'm Wrong: Using Non-Experts to Repair Reinforcement Learning Policies.
Sanne van WaverenChristian PekJana TumovaIolanda LeitePublished in: HRI (2022)
Keyphrases
- reinforcement learning
- optimal policy
- policy search
- control policies
- function approximation
- reinforcement learning agents
- markov decision process
- reward function
- state space
- markov decision processes
- fitted q iteration
- reinforcement learning algorithms
- partially observable markov decision processes
- markov decision problems
- dynamic programming
- learning algorithm
- temporal difference
- control policy
- decision problems
- multi agent reinforcement learning
- decentralized control
- policy gradient methods
- robotic control
- reinforcement learning methods
- machine learning
- average cost
- infinite horizon
- human experts
- learning classifier systems
- neural network
- multi agent
- learning process
- reward shaping
- macro actions
- domain knowledge
- hierarchical reinforcement learning
- policy iteration
- management policies
- long run
- action selection
- partially observable
- action space