ePrints

Reinforcement learning in linear embedding space unlocks generalizable control across soft robot configurations

Lookup NU author(s): Professor Wei PanORCiD

Downloads

• Published version [.pdf]

Licence

This work is licensed under a Creative Commons Attribution-NonCommercial-NoDerivatives 4.0 International License (CC BY-NC-ND).

Abstract

© The Author(s) 2026.Soft-bodied organisms exhibit prominent morphological adaptability, dynamically reconfiguring shape and stiffness to achieve versatile behaviors. Inspired by these systems, soft robots with diverse morphologies have emerged, yet a unified control framework that rapidly adapts across configurations remains elusive. Here, we introduce a generalizable control system that enables rapid cross-configuration adaptation via reinforcement learning in a shared linear Koopman embedding space. By encoding robot dynamics into this embedding space, our method decouples control policies from specific morphologies, allowing real-time, model-free policy adaptation without retraining from scratch. We validate our system across 33 distinct robot configurations. Our system achieves a 75 × reduction in transfer samples across configurations, while sustaining robust performance under high-speed motion, heavy payloads, and multiactuator faults, and achieving real-world skills previously unattainable in soft robotics. This work establishes an adaptable control framework for diverse soft robot configurations and may offer insights for generalizable control in complex physical systems.

Publication metadata

Author(s): Zhang X, Li C, Mo H, Jiang Y, Cao W, Xu X, Jiang W, Bing Z, Yang Y, Li X, Yang Y, Lu H, Zeng L-L, Knoll A, Hu D, Wen L, Pan W

Publication type: Article

Publication status: Published

Journal: Nature Communications

Year: 2026

Volume: 17

Issue: 1

Online publication date: 08/06/2026

Acceptance date: 14/04/2026

Date deposited: 04/08/2026

ISSN (electronic): 2041-1723

Publisher: Nature Research

URL: https://doi.org/10.1038/s41467-026-72491-9

DOI: 10.1038/s41467-026-72491-9

Data Access Statement: All data needed to evaluate the conclusions in the paper are presented in the paper and in the Supplementary Information. Movies and source data are provided with the submission. Source data are provided with this paper. All codes for the Koopman embedding pretraining process and the online RL algorithm are provided with the submission52 via the GitHub repository https://doi.org/10.5281/zenodo.18901936

PubMed id: 42259801

Altmetrics

Funding

Share