Context-Continuous Preference Learning for Exoskeleton Personalization

Sunin Baek1, Sungwoo Park2,*, Daekyum Kim1,3,*

1 Korea University, School of Mechanical Engineering
2 WIRobotics, Innovation Lab
3 Korea University, School of Smart Mobility
* Corresponding authors
Paper arXiv
Context-Continuous Preference Learning for Exoskeleton Personalization

Abstract

Exoskeleton personalization across multiple operating conditions is limited by the time and physical effort needed to collect user feedback. This study investigates whether user preference landscapes vary smoothly across operating contexts and when such continuity can support learning with fewer observations. CCPL models preference jointly over assistance parameters and context so that information from nearby conditions can contribute to context-specific estimates. Simulation studies and retrospective analyses of ankle and elbow exoskeleton preference data from nine healthy adults show that context-aware sharing can improve preference reconstruction and held-out response prediction under limited feedback. The simulations also show that inappropriate sharing can produce negative transfer when preferences are not sufficiently continuous across contexts. These results motivate context continuity as a useful structure for data-efficient exoskeleton personalization, while online benefits in human personalization remain to be established.

BibTeX

@article{baek2026ccpl,
  title={Context-Continuous Preference Learning for Exoskeleton Personalization},
  author={Baek, Sunin and Park, Sungwoo and Kim, Daekyum},
  journal={arXiv preprint arXiv:2609.28427},
  year={2026}
}