Core Concepts
The author proposes a method using causal trees to automatically generate regions of different task difficulty levels based on individual performance, outperforming previous approaches. The main thesis is to develop a technique that estimates task difficulty based on features without the need for robots to exert forces on users.
Abstract
Using causal trees, this study aims to estimate personalized task difficulty levels for post-stroke individuals. By analyzing user performance in reaching tasks, the method provides insights into individual limitations and informs personalized rehabilitation programs. The approach surpasses traditional methods by offering better explanations for user performance variance and enabling adaptive training programs crucial for recovery post-stroke.
Stats
Table I: Performance of Predicting Personalized Difficulty
Causal Tree: avg. r2 .656, std. err. r2 .007, p-value -
Random Forest: avg. r2 .635, std. err. r2 .006, p-value .036
Neural Network: avg. r2 .634, std. err. r2 .005, p-value .023
Support Vector Machine: avg. r2 .598, std. err. r2 .010, p-value < .001
5-Nearest Neighbors: avg. r2 .572, std. err. r2 .007, p-value < .001
Decision Tree: avg. r2 .588, std. err.r2 .007, p-value < .001
Quotes
"We propose a method that automatically generates regions of different task difficulty levels based on an individual’s performance."
"Our technique applies to holistic tasks and does not require the robot to exert forces on the user."
"Causal trees significantly outperform all baselines when estimating personalized task difficulty."