PHABS: A Handheld Haptic Device for Force-Annotated Bimanual Demonstration Data

Calvin Joyce, Michael C. Yip

Keywords: haptic feedback, bimanual manipulation, teleoperation, force-annotated data, demonstration collection, imitation learning, handheld device, force feedback, dexterous manipulation, robot learning

TL;DR: PHABS is an in-development handheld bimanual haptic device that captures synchronized pose and pinching-force data during two-handed manipulation, serving as a force-rich anchor to complement force-blind human-video and simulation datasets.

Abstract: Human video and simulation are increasingly used to sidestep the bottlenecks of teleoperated data collection. Yet, none of these sources, video, simulation, or existing handheld teleoperation, reliably captures the contact forces that define skilled manipulation. We argue that small, force-rich teleoperation data can serve as a force-anchoring modality that complements larger, force-blind video and simulation corpora. We present an in-development device, PHABS (Portable Haptic Assisted Bimanual System). This handheld bimanual device captures synchronized position and pinching-force data during two-handed manipulation while rendering haptic feedback to the operator via capstan-driven actuators. The design targets two-handed manipulation of the same or related objects with fine pinching-force control on each hand, a class underrepresented in existing datasets because it is difficult to capture either from third-person video or from force-blind teleop. A pilot study (n=3) on stiffness discrimination and bimanual pick-and-place indicates that closed-loop force feedback improves task success and operator confidence relative to a position-only baseline. We position PHABS as an in-progress hardware infrastructure intended for collecting a force-annotated anchor dataset, and we report both a working pinching subsystem and lessons from a lateral-force subsystem still under revision.

Email Sharing: We authorize the sharing of all author emails with Program Chairs.

Data Release: We authorize the release of our submission and author names to the public in the event of acceptance.

Submission Number: 18