Robotic Compliant Object Prying Using Vision and Force Guided Diffusion Policy скачать в хорошем качестве

Robotic Compliant Object Prying Using Vision and Force Guided Diffusion Policy

The increasing use of batteries in electric vehicles and consumer products highlights the need for efficient recycling solutions. These devices often combine rigid and compliant components, making robotic disassembly crucial for scalable recycling. Diffusion policy has shown promise for learning robotic skills, but contact-rich tasks require integrating force feedback. We apply diffusion policy with vision and force to a compliant object prying task. To prevent force data from being overshadowed by high-dimensional image data, we propose a method to effectively integrate them. Our approach achieves 96% success on a battery prying task, a 57% improvement over vision-only baselines, and demonstrates zero-shot transfer to unseen objects and batteries. More details can be found in our project website: https://rros-lab.github.io/diffusion-...