@inproceedings{45b46e1118bd475b903b997a87ed1d40,
title = "Physically-Informed Multi-Modal Diffusion for Robotic Plug-and-Unplug Manipulation",
abstract = "Plug-and-unplug manipulation is a representative contact-rich task that requires precise coordination of visual, depth, and force feedback. This paper presents a physically-informed multi-modal diffusion framework for robotic plug- and-unplug manipulation. The proposed method integrates RGB-D vision, force/torque sensing, and proprioceptive data into a unified latent space through a variational fusion network. Conditioned on this representation, a diffusion model generates continuous action trajectories consistent with contact dynamics. An action library is constructed based on generated trajectories, organized by contact semantics, to enable skill retrieval and transfer. Preliminary results in simulation validate the model{\textquoteright}s ability to encode multi-modal signals and generate physically plausible manipulation actions. Future work will focus on real-robot experiments and robustness evaluation.",
keywords = "control, docking, multi-modal diffusion, Plug- and -unplug manipulation",
author = "Chenyu Zhang and Meng Chen and Yuying Liang",
note = "Publisher Copyright: {\textcopyright} 2026 The Authors.; 9th International Conference on Aeronautical, Aerospace and Mechanical Engineering, AAME 2026 ; Conference date: 27-03-2026 Through 29-03-2026",
year = "2026",
month = aug,
day = "17",
doi = "10.3233/ATDE260548",
language = "英语",
series = "Advances in Transdisciplinary Engineering",
publisher = "IOS Press BV",
pages = "530--540",
editor = "Mo, \{John P.T.\}",
booktitle = "Moving Integrated Product Development to Service Clouds in the Global Economy - Proceedings of the 21st ISPE Inc. International Conference on Concurrent Engineering, CE 2014",
address = "荷兰",
}