Unifying Unsupervised and Offline RL for Fast Adaptation Using World Models

Keyphrases

• Fast Adaptation 100%
• World Model 100%
• Unsupervised RL 100%
• Offline RL 100%
• Adaptation 50%
• Online Performance 50%
• Robotics 50%
• Possible Sources 50%
• Effective Method 50%
• Recent Approaches 50%
• Control Plan 50%
• Deep Reinforcement Learning (deep RL) 50%
• Exploration Model 50%
• Model Predictive Control 50%
• Generative Models 50%
• Reward Signal 50%
• Offline Data 50%
• Task Adaptation 50%
• Unsupervised Pre-training 50%
• Online Adaptation 50%
• Data-driven Predictive Control 50%
• New Tasks 50%
• Expert Demonstration 50%
• Unsupervised Reinforcement Learning 50%
• Data Efficiency 50%
• Unsupervised Training 50%
• Exploration Data 50%
• Different Types of Data 50%
• Data Generalization 50%

Computer Science

• Data Stream 100%
• Predictive Model 100%
• Reinforcement Learning 100%
• Effective Method 100%
• Deep Reinforcement Learning 100%
• Generative Model 100%
• Robotics 100%
• Data Exploration 100%