This paper presents a comprehensive study on the generalizability of task models for upper-limb exoskeletons. The researchers focused on six different regression algorithms - Locally Weighted Projection Regression (LWPR), K-Nearest Neighbours (KNN), Support Vector Regression (SVR), eXtreme Gradient Boosting (XGBoost), Multi Layer Perceptron (MLP), and Gaussian Process Regression (GPR) - and evaluated their performance on six distinct tasks: Horizontal, Vertical, diagonal from Left leg to Right eye, diagonal from Right leg to Left eye, Eating, and Pushing.
The key findings are:
The task models performed along the horizontal plane and decision tree-based algorithms (e.g., XGBoost) exhibited superior generalizability compared to other models.
The generalizability of the task models can be ordered as: Horizontal, diagonal from Right leg to Left eye, Eating, Pushing, Vertical, and diagonal from Left leg to Right eye.
The average training times for the regression algorithms ranged from 0.003 seconds for LWPR to 0.247 seconds for MLP, providing insights into the computational efficiency of the models.
These insights can guide the development of more generalizable and adaptable assistive robotics systems, reducing the need for extensive data collection and retraining for each new task or user interaction.
To Another Language
from source content
arxiv.org
Key Insights Distilled From
by Hamid Osooli... at arxiv.org 05-07-2024
https://arxiv.org/pdf/2405.02492.pdfDeeper Inquiries