CART-MPC: Coordinating Assistive Devices for Robot-Assisted Transferring with Multi-Agent Model Predictive Control

TL;DR

This work tackles bed-to-wheelchair transferring with limited robot payload by instrumenting assistive devices (Hoyer sling and wheelchair) to act as cooperative agents. It introduces CART-MPC, a turn-taking multi-agent model predictive control framework that coordinates a robot arm and a rotating sling bar using a learned keypoint-space dynamics model and a knot-theory–inspired cost based on the $linking\ number$, with neural amortization for fast planning. The approach generalizes across hook shapes, sling materials, and care-recipient body shapes in simulation and demonstrates zero-shot sim-to-real transfer to a manikin, highlighting practical potential for autonomous transfer. By integrating deformable-object manipulation, multi-agent coordination, and sim-to-real transfer, CART-MPC advances autonomous caregiving robotics toward safer, more capable bed-to-chair transferring.

Abstract

Bed-to-wheelchair transferring is a ubiquitous activity of daily living (ADL), but especially challenging for caregiving robots with limited payloads. We develop a novel algorithm that leverages the presence of other assistive devices: a Hoyer sling and a wheelchair for coarse manipulation of heavy loads, alongside a robot arm for fine-grained manipulation of deformable objects (Hoyer sling straps). We instrument the Hoyer sling and wheelchair with actuators and sensors so that they can become intelligent agents in the algorithm. We then focus on one subtask of the transferring ADL -- tying Hoyer sling straps to the sling bar -- that exemplifies the challenges of transfer: multi-agent planning, deformable object manipulation, and generalization to varying hook shapes, sling materials, and care recipient bodies. To address these challenges, we propose CART-MPC, a novel algorithm based on turn-taking multi-agent model predictive control that uses a learned neural dynamics model for a keypoint-based representation of the deformable Hoyer sling strap, and a novel cost function that leverages linking numbers from knot theory and neural amortization to accelerate inference. We validate it in both RCareWorld simulation and real-world environments. In simulation, CART-MPC successfully generalizes across diverse hook designs, sling materials, and care recipient body shapes. In the real world, we show zero-shot sim-to-real generalization capabilities to tie deformable Hoyer sling straps on a sling bar towards transferring a manikin from a hospital bed to a wheelchair. See our website for supplementary materials: https://emprise.cs.cornell.edu/cart-mpc/.

Figures (6)

• Figure 1: Bed-to-wheelchair Transferring: Human caregivers use assistive devices to perform transferring. For example, the caregiver in the figure (upper row) uses a Hoyer sling to move the care recipient from the hospital bed to the wheelchair. Inspired by this, we use an instrumented Hoyer sling along with a caregiving robot to perform the transferring task (lower row), where the robot performs fine manipulation, and the assistive devices take the heavy loads.
• Figure 2: Instrumented assistive devices: We added sensors and actuators to a Hoyer sling and a commercial powered wheelchair with a robot arm.
• Figure 3: (a) CART-MPC: We propose a turn-taking multi-agent algorithm to coordinate the robot and the Hoyer sling bar during strap fastening. The algorithm leverages a dynamics model for the strap and a neural cost function. See text for details. (b) Neural linking number: We visualize the average neural linking number across five strap-tying trials. The neural linking number gradually increases from 0 to 1 as the robot ties the strap onto the hook.
• Figure 4: Setup and results for evaluation of CART-MPC in RCareWorld: We evaluate our method using the setup on the left with various Hook shapes (H), Care Recipients (CR) with various body shapes, and Sling Materials (M) with three levels of compliance. We show the results $S_1$ in (b) comparing CART-MPC with a single-agent MPC with a passive sling bar. The results suggest that CART-MPC using multi-agent performs better than a single-agent baseline method. See Table \ref{['tab:main-results']} for full results.
• Figure 5: CART-MPC executed sequence in RCareWorld: We demonstrate one executed sequence with H1, M1, and CR3. In this trial, the robot and the sling bar collaboratively fasten the 4 straps to the sling bar hooks.