Structure-Preserving Model Order Reduction of Slender Soft Robots Via Autoencoder-Parameterized Strain
Abdulaziz Y. Alkayas, Anup Teejo Mathew, Daniel Feliu, Yahya Zweiri, Thomas George Thuruthel, Federico Renda
AI summary
Problem
High-dimensional soft robot models are computationally expensive to simulate, and existing reduced-order techniques struggle to efficiently capture complex, nonlinear strain fields without sacrificing accuracy or physical interpretability.
Approach
Autoencoders learn low-dimensional nonlinear strain parameterizations from high-order simulation data, which are then integrated into the Geometric Variable Strain framework to construct a structure-preserving reduced-order model.
Key results
- AE-based ROMs achieve lower strain errors than POD for equivalent degrees of freedom
- Computational speed-ups of up to 3.9x over POD and over 15x over high-order models
- Autoencoders automatically discover the intrinsic low-dimensional manifolds of soft robot dynamics
- Experimental validation demonstrates near-identical manipulator behavior using only a single degree of freedom
Why it matters
Enables real-time simulation and control of complex soft robots by drastically reducing computational cost while preserving physical model interpretability.
Abstract
While soft robots offer advantages in adaptability and safe interaction, their modeling remains challenging. This letter presents a novel, data-driven approach for model order reduction of slender soft robots using autoencoder-parameterized strain within the Geometric Variable Strain (GVS) framework. We employ autoencoders (AEs) to learn low-dimensional strain parameterizations from data to construct reduced-order models (ROMs), preserving the Lagrangian structure of the system while significantly reducing the degrees of freedom. Our comparative analysis demonstrates that AE-based ROMs consistently outper- form proper orthogonal decomposition (POD) approaches, achiev- ing lower errors for equivalent degrees of freedom across mul- tiple test cases. Additionally, we demonstrate that our proposed approach achieves computational speed-ups over the high-order models (HOMs) in all cases, and outperforms the POD-based ROM in scenarios where accuracy is matched. We highlight the intrinsic dimensionality discovery capabilities of autoencoders, revealing that HOM often operate in lower-dimensional nonlinear mani- folds. Through both simulation and experimental validation on a cable-actuated soft manipulator, we demonstrate the effectiveness of our approach, achieving near-identical behavior with just a single degree of freedom. This structure-preserving method offers Received 25 March 2025; accepted 17 August 2025. Date of publication 4 September 2025; date of current version 15 September 2025. This article was recommended for publication by Associate Editor F. Chen and Editor C. Laschi upon evaluation of the reviewers’ comments. This work was supported in part by the US Office of Naval Research Global under Grant N62909-21-1-2033 and in part by Khalifa University under Award RIG-2023-048 and Award RC1-2018- KUCARS. (Corresponding author: Abdulaziz Y. Alkayas.) Abdulaziz Y. Alkayas is with the Department of Mechanical and Nuclear Engineering, Khalifa University, Abu Dhabi 52799, UAE, and also with the Department of Computer Science, University College London, WC1E 6BT London, U.K. (e-mail: abdulaziz.alkayas@ku.ac.ae). Anup Teejo Mathew and Federico Renda are with the Department of Me- chanical and Nuclear Engineering, Khalifa University, Abu Dhabi 52799, UAE, and also with the Khalifa University Center for Autonomous Robotic Systems (KUCARS), Khalifa University, Abu Dhabi 127788, UAE (e-mail: anup.mathew@ku.ac.ae; federico.renda@ku.ac.ae). Daniel Feliu-Talegon is with the Department of Mechanical and Nuclear Engineering, Khalifa University, Abu Dhabi 52799, UAE, and also with the Department of Cognitive Robotics, Delft University of Technology, 2628 CN Delft, Netherlands (e-mail: d.feliutalegon@tudelft.nl). Yahya Zweiri is with the Khalifa University Center for Autonomous Robotic Systems (KUCARS), Khalifa University, Abu Dhabi 127788, UAE, and with the Department of Aerospace Engineering, Khalifa University, Abu Dhabi 52799, UAE, and also with the Advanced Research and Innovation Center (ARIC), Khalifa University, Abu Dhabi 52799, UAE (e-mail: yahya.zweiri@ku.ac.ae). Thomas George Thuruthel is with the Department of Computer Science, University College London, WC1E 6BT London, U.K. (e-mail: t.thuruthel@ucl. ac.uk). This article has supplementary downloadable material available at https://doi.org/10.1109/LRA.2025.3606389, provided by the authors. Digital Object Identifier 10.1109/LRA.2025.3606389 significant reductions in the system degrees of freedom and com- putational effort while maintaining physical model interpretability, offering a promising direction for soft robot modeling and control.