Modeling Personalized Difficulty of Rehabilitation Exercises Using Causal Trees

TL;DR

This work addresses how to personalize rehabilitation exercise difficulty to accommodate diverse stroke survivors. It introduces a causal-tree framework that combines nominal difficulty measured from neurotypical participants with functional difficulty from post-stroke users, formalized as $\tau(\vec{x}) = \mathbb{E}[Y^{(1)} - Y^{(0)} | X=\vec{x}]$, and estimates it via a causal decision tree that partitions the task space. Across a reaching-task study with neurotypical and post-stroke participants, the approach achieves lower mean-squared error and higher explained variance ($r^2$) than baseline models, while providing interpretable visualizations of difficulty regions. The method enables practitioners to tailor rehabilitation tasks, improve motivation, and communicate difficulty rationale to patients and clinicians, with potential applicability to a broad class of adaptive rehabilitation activities. Future work includes multi-outcome difficulty estimation and extension to additional tasks beyond reaching.

Abstract

Rehabilitation robots are often used in game-like interactions for rehabilitation to increase a person's motivation to complete rehabilitation exercises. By adjusting exercise difficulty for a specific user throughout the exercise interaction, robots can maximize both the user's rehabilitation outcomes and the their motivation throughout the exercise. Previous approaches have assumed exercises have generic difficulty values that apply to all users equally, however, we identified that stroke survivors have varied and unique perceptions of exercise difficulty. For example, some stroke survivors found reaching vertically more difficult than reaching farther but lower while others found reaching farther more challenging than reaching vertically. In this paper, we formulate a causal tree-based method to calculate exercise difficulty based on the user's performance. We find that this approach accurately models exercise difficulty and provides a readily interpretable model of why that exercise is difficult for both users and caretakers.

Figures (5)

• Figure 1: Multiple views of two users' personalized difficulty scores. Lighter colors indicate regions of higher difficulty, determined by statistically significant increases in individual reach times from baseline values.
• Figure 2: A post-stroke participant performing a reaching exercise. The Socially Assistive Robot (left) provided instructions and feedback. The robot arm (right) moved the reaching target in front of the user. The user reached to the button 100 times using their more affected side.
• Figure 3: Participants' workspace. The arm moved the target to different points within the blue-shaded region to estimate difficulty.
• Figure 4: Visual comparison of methods. We show the user's ground truth difference from the neurotypical population (a). Independently estimating the stroke survivor and neurotypical data leads to homogenous difficulty estimates over the workspace (b). Using causal trees instead provides distinctive regions of differences in difficulty that are robust to outliers (c).
• Figure 5: Tree interpretation of the difficulty map for PID 33.