Scaling data-driven robotics with reward sketching and batch reinforcement learning.
Serkan CabiSergio Gómez ColmenarejoAlexander NovikovKsenia KonyushkovaScott E. ReedRae JeongKonrad ZolnaYusuf AytarDavid BuddenMel VeceríkOleg SushkovDavid BarkerJonathan ScholzMisha DenilNando de FreitasZiyu WangPublished in: Robotics: Science and Systems (2020)
Keyphrases
- reinforcement learning
- data driven
- perception action
- artificial intelligence
- eligibility traces
- batch mode
- robot control
- reinforcement learning algorithms
- reward function
- function approximation
- computer vision
- markov decision processes
- machine learning
- state space
- batch processing
- real robot
- model free
- partially observable environments
- optimal policy
- action selection
- learning process
- reinforcement learning methods
- reward shaping
- industrial robots
- state action
- learning agent
- average reward
- markov decision process
- partially observable
- robotic systems
- learning problems
- control policy
- policy search
- supervised learning
- machine intelligence