Carnegie Mellon University Robotics Institute转发了
Low-cost teleoperation systems have democratized robot data collection, but they lack any force feedback, making it challenging to teleoperate contact-rich tasks. Meanwhile, many robot arms provide force information — a critical yet underutilized modality in robot learning. We introduce: 1. ?? A low-cost, ?force-feedback-enabled teleoperation system.? 2. ?? Force-Attending Curriculum Training (FACTR) uses force to improve generalization in complex, contact-rich tasks. ??Our teleop system (~$1K per leader arm) lets users intuitively feel forces experienced by robots, without additional sensor hardware. This can be done for many arms. We also integrate gravity compensation, redundancy resolution, etc. in the teleop leader arm, augmenting the teleop experience. ??Behavior cloning policies often overfit to vision input and ignore force when added naively, limiting performance in contact-rich tasks. We propose FACTR, a curriculum that prevents overfitting to vision and guides the policy to properly attend to force. FACTR policies generalize to unseen objects significantly better than baseline force policies without FACTR. FACTR is easy to set up on existing systems. Both the hardware and code of FACTR will be open-sourced soon! This work is my first PhD project at Carnegie Mellon University with co-lead Yulong Li as well as Kenny Shaw, Tony Tao, Russ Salakhutdinov, and Deepak Pathak. Webpage: https://lnkd.in/dSTyavk4 Paper: https://lnkd.in/d4iYjgmp