About this document

Robotic Offline RL From Internet Videos Via Value-Function Pre-Training by ahad is a document available to read on EtoBox.

The paper presents V-PTR, a system that leverages large-scale human video datasets to enhance robotic offline reinforcement learning (RL) by learning value functions through temporal-difference learning. This approach addresses the challenge of integrating action-free video data into RL methods, resulting in improved generalization and robustness in manipulation tasks on a real robot. Experimental results demonstrate that V-PTR significantly outperforms prior methods, showcasing the effectiveness of using v

Author

ahad

Language

EN