Claire Chen

@ChaoyiPan Thanks! 😀 We compute target fingertip positions using measured contact forces. When tracked with a Cartesian impedance controller, these actions allow the robot to generate the right motion and forces to complete the task.

@clairelchen Cool! For other LfD work, force seems to be achieved by extra position offset from human demonstrators. But now, it seems that the demonstration also includes external force from humans. Is there any lower level controller to achieve the desired force without human help?

Contact-rich dexterous manipulation, like opening an AirPods case or unscrewing a nut, requires a robot to apply the right forces at the right moments. Our new work DexForce leverages force sensing to get actions that enable robot hands to perform these contact-rich tasks. (1/6)

DexForce features coin-sized 6-axis force/torque sensors, called CoinFT, designed by @_Hojung_Choi from @bdmlstanford. It was great to work with Hojung, as well as @zhongchunO3O, Mark Cutkosky, and @leto__jean. Check out our project page for more details: clairelc.github.io/dexforce.githu…

We also ablate including force data in policy observations. We find that while force observations never hurt policy performance, they help the most for tasks that require advanced precision and coordination, like opening an AirPods case, unscrewing a nut, and sliding cube. (5/6)