Reinforcement Learning with Action-Free Pre-Training from Videos.
Younggyo SeoKimin LeeStephen L. JamesPieter AbbeelPublished in: ICML (2022)
Keyphrases
- reinforcement learning
- human actions
- partially observable domains
- video frames
- action selection
- action classification
- video sequences
- supervised learning
- training set
- learning algorithm
- spatio temporal
- recognition of human actions
- training samples
- action recognition
- recognizing human actions
- human activities
- multi agent
- action space
- machine learning
- reward shaping
- view invariant
- fully labeled
- video database
- training process
- video content
- markov decision processes
- training examples
- state space
- video analysis
- video clips
- recognizing actions
- moving objects