GRACE: Generalizing Robot-Assisted Caregiving with User Functionality Embeddings

TL;DR

A neural model is developed that learns to embed functional assessment scores into a latent representation of the user's physical function that enables the robot to provide personalized and effective assistance while improving the user's agency in action.

Abstract

Robot caregiving should be personalized to meet the diverse needs of care recipients -- assisting with tasks as needed, while taking user agency in action into account. In physical tasks such as handover, bathing, dressing, and rehabilitation, a key aspect of this diversity is the functional range of motion (fROM), which can vary significantly between individuals. In this work, we learn to predict personalized fROM as a way to generalize robot decision-making in a wide range of caregiving tasks. We propose a novel data-driven method for predicting personalized fROM using functional assessment scores from occupational therapy. We develop a neural model that learns to embed functional assessment scores into a latent representation of the user's physical function. The model is trained using motion capture data collected from users with emulated mobility limitations. After training, the model predicts personalized fROM for new users without motion capture. Through simulated experiments and a real-robot user study, we show that the personalized fROM predictions from our model enable the robot to provide personalized and effective assistance while improving the user's agency in action. See our website for more visualizations: https://emprise.cs.cornell.edu/grace/.

Figures (5)

• Figure 1: We present GRACE, a method for generalizing robot-assisted caregiving by creating personalized user models through predicting functional range of motion (fROM) from functional assessments.
• Figure 2: Method overview. Using functional assessments scored by occupational therapists (OT) and ground-truth motion capture data, we train a neural model to predict fROM for new users. See text for details.
• Figure 3: Left: Collecting data for a subject with emulated mobility limitations using ARAT. Center and Right: We collected functional assessments and fROM data (tracked by motion capture) with 11 subjects, each emulating four conditions. $U_1$, $F_1$, $W_1$ are anchor points on the upper arm, forearm, and waist.
• Figure 4: Evaluation compared with two baselines in four settings shows that GRACE generalizes to unseen users and conditions.
• Figure 5: Top Left: Visualizations of the four environments in our simulation experiments. Bottom Left: Simulation evaluations show that GRACE achieves better performance balancing task success and agency in action compared to heuristic baselines. Right: User study results show that GRACE improves user's sense of agency, physical effort, and safety while providing effective assistance.