Weight Regression for a Generalized Motion Primitive Formulation in Cooperative Hand Placement Tasks with Upper-Limb Prostheses
Hongjun Cai, Rebecca J. Greene, Christopher L. Hunt, Nitish V. Thakor
Abstract
Recent years have seen a growing interest in the development of shared control strategies for upper limb prostheses. In this work, we take a critical step towards developing transhumeral devices by proposing a biomimetic control strategy for cooperative hand placement. This is achieved through a novel adaptation of Dynamic Movement Primitives (DMPs), enabling the generation of smooth trajectories from a rest position to arbitrary points within a user’s reach space. Our method revolves around a key observation that DMP forcing-function weights can be modeled (p < 0.05) for ≥90% of values as a simple function of Cartesian position, achieving median R<sup xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">2</sup> values of 0.63, 0.43, and 0.02 (horizontal, vertical, and depth). Validation on 519 trajectories via 5-fold cross-validation showed significant improvements (p < 0.01) over Extended DMP and kernel-based methods. Real-time human-in-the-loop experiments revealed a median minimum cumulative-distance deviation of 0.0733 m (8.5% error) motion with a prosthesis as compared with an intact limb. To our knowledge, this is the first study to explore shared control for transhumeral prostheses, and our observations on human motion modeling may inspire future Learning-from-Demonstration studies.
BibTeX
@inproceedings{iros2025_weightregression,
title = {Weight Regression for a Generalized Motion Primitive Formulation in Cooperative Hand Placement Tasks with Upper-Limb Prostheses},
author = {Hongjun Cai and Rebecca J. Greene and Christopher L. Hunt and Nitish V. Thakor},
booktitle = {IROS 2025},
year = {2025}
}