Stylized Offline Reinforcement Learning: Extracting Diverse High-Quality Behaviors from Heterogeneous Datasets.
Yihuan MaoChengjie WuXi ChenHao HuJi JiangTianze ZhouTangjie LvChangjie FanZhipeng HuYi WuYujing HuChongjie ZhangPublished in: ICLR (2024)
Keyphrases
- high quality
- reinforcement learning
- low quality
- real time
- real world
- ground truth
- highly heterogeneous
- optimal policy
- real robot
- state space
- function approximation
- benchmark datasets
- multi agent
- data sets
- behavior analysis
- temporal difference
- action selection
- higher quality
- image quality
- high resolution
- learning algorithm
- model free
- automatically extracted
- action space
- behavior recognition
- transition model
- robotic control