RABBIT: A Robot-Assisted Bed Bathing System with Multimodal Perception and Integrated Compliance

TL;DR

RABBIT introduces a robot-assisted bed bathing system that fuses RGB-T multimodal perception with dual software and hardware compliance to execute washing, rinsing, and drying autonomously. It trains segmentation models on a synthetic SBP dataset, deploys caregiver-inspired motion primitives, and uses Scrubby end-effector with gain-scheduled impedance control for safe pHRI. A user study with 12 participants, including one with severe mobility limitations, demonstrates high comfort and performance, indicating potential to reduce caregiver burden and improve hygiene for mobility-impaired individuals. The work provides open-source tools and data to advance perception and compliant control in assistive robotics.

Abstract

This paper introduces RABBIT, a novel robot-assisted bed bathing system designed to address the growing need for assistive technologies in personal hygiene tasks. It combines multimodal perception and dual (software and hardware) compliance to perform safe and comfortable physical human-robot interaction. Using RGB and thermal imaging to segment dry, soapy, and wet skin regions accurately, RABBIT can effectively execute washing, rinsing, and drying tasks in line with expert caregiving practices. Our system includes custom-designed motion primitives inspired by human caregiving techniques, and a novel compliant end-effector called Scrubby, optimized for gentle and effective interactions. We conducted a user study with 12 participants, including one participant with severe mobility limitations, demonstrating the system's effectiveness and perceived comfort. Supplementary material and videos can be found on our website https://emprise.cs.cornell.edu/rabbit.

Figures (8)

• Figure 1: RABBIT is a robot-assisted bed bathing system informed by expert caregiving practices. By effectively executing the three bathing tasks of washing, rinsing, and drying, we achieve a fully autonomous cleaning trial on the forearm of a user with severe mobility limitations.
• Figure 2: The key elements of our proposed system. The system first captures a photo of the region it wants to bathe, and segments it as soapy, wet, or dry. Based on the task a path is planned and then the controller executes the plan.
• Figure 3: Top shows user and robot setup, sensors, and bathing tools for each bathing task. The bottom shows the enhanced manikin arm data collection setup.
• Figure 4: RGB and thermal images of Enhanced Manikin look realistic and closely resemble features of a human limb.
• Figure 5: Qualitative comparison of segmentation models trained on SBP dataset.