Reinforcement Learning with Action-Free Pre-Training from Videos.
Younggyo SeoKimin LeeStephen JamesPieter AbbeelPublished in: CoRR (2022)
Keyphrases
- reinforcement learning
- human actions
- action selection
- function approximation
- video sequences
- supervised learning
- recognizing human actions
- reward shaping
- partially observable domains
- video frames
- action recognition
- multi agent
- learning algorithm
- training set
- action classification
- video surveillance
- optimal control
- training process
- reinforcement learning algorithms
- state action
- recognition of human actions
- recognizing actions
- neural network
- fully labeled
- continuous state
- key frames
- dynamic programming
- machine learning