Robot learning and perception for contact-rich manipulation

Abstract: Object rearrangement will become an important skill for collaborative robots as they increasingly operate in
unstructured environments. Rearrangement involves grasping, goal-conditioned navigation, and placement. In this talk, I will present our latest work, which focuses on placement. We learn a haptics-based RL policy in
simulation for inserting household objects into slots, and then transfer it to a real robot without fine-tuning or adaptation. I will also present some insights we gained along the way about which factors influence sim- …

abstract become collaborative collaborative robots environments fine-tuning influence insights learn navigation objects perception policy robot robots sim simulation talk transfer unstructured work