guozhong luo

No description.

A research system that trains robots to perform multi-step physical tasks by generating predicted images of future workspace states as visual subgoals, achieving 100 percent success on five industrial manipulation benchmarks versus 80 percent for the baseline.

Official implementation of COST-HIL: Cost-Aware Human-in-the-Loop Reinforcement Learning for Real-World Dexterous Manipulation. This repo contains source codes, experiment logs, real-robot test scripts and supplementary materials for our CVPR submission.

No description.