Robust Body Exposure (RoBE): A Graph-based Dynamics Modeling Approach to Manipulating Blankets over People

TL;DR

This work tackles automated, targeted bedding manipulation for in-bed care by learning a graph-based cloth dynamics model and performing run-time optimization to uncover targeted body parts under a blanket. RoBE uses a 2D cloth graph and a four-layer graph neural simulator to predict cloth motion, then optimizes actions with CMA-ES against a body-part coverage objective that penalizes exposing non-target regions. Across simulation, a real-world manikin, and a human study, RoBE demonstrates superior generalization to variations in body shape, pose, and blanket configuration compared to a geometric baseline and PPO-based RL, validating a robust, data-efficient approach for caregiving robotics. The method enables flexible, privacy-preserving bedside assistance by adapting to real-world variations without retraining, highlighting practical potential for robot-assisted care in hospital and home settings.

Abstract

Robotic caregivers could potentially improve the quality of life of many who require physical assistance. However, in order to assist individuals who are lying in bed, robots must be capable of dealing with a significant obstacle: the blanket or sheet that will almost always cover the person's body. We propose a method for targeted bedding manipulation over people lying supine in bed where we first learn a model of the cloth's dynamics. Then, we optimize over this model to uncover a given target limb using information about human body shape and pose that only needs to be provided at run-time. We show how this approach enables greater robustness to variation relative to geometric and reinforcement learning baselines via a number of generalization evaluations in simulation and in the real world. We further evaluate our approach in a human study with 12 participants where we demonstrate that a mobile manipulator can adapt to real variation in human body shape, size, pose, and blanket configuration to uncover target body parts without exposing the rest of the body. Source code and supplementary materials are available online.

Figures (9)

• Figure 1: Using RoBE to expose both lower legs of a participant in the real world using a Stretch RE1 mobile manipulator.
• Figure 2: Overview of our approach: A) We generate a set of body points from an observation of a person's pose. B) We cover them with a blanket, then capture and process an initial cloth point cloud for composition of a graph. C) Given an input graph, we run an optimization loop over the dynamics model to find an action that uncovers a target body part. D) We generate an initial prediction of performance before executing the best action in the real world.
• Figure 3: A simulation rollout: 1) We initialize a human model above the bed. 2) After the human model settles on the bed, we initialize a blanket above them. 3) We drop the blanket onto the human model and the end effector picks up the blanket at the grasp location. 4) The blanket is lifted upwards before being moved to the release location.
• Figure 4: Capturing human body shape parameters and pose in the real world: 1) Measuring each of the nine body segments, 2) Adjusting the estimated position of the 14 joints in observation $\bm{s}$ to better match the ground truth pose, 3) visualization of all the virtual points generated along the body.
• Figure 5: Comparison of the action space used in simulation, $A$, and the action space used in the real world implementation, $A_{RW}$.