Details
| Original language | English |
|---|---|
| Pages (from-to) | 619 - 636 |
| Number of pages | 18 |
| Journal | IEEE Transactions on Robotics |
| Volume | 42 |
| Early online date | 12 Nov 2025 |
| Publication status | Published - 16 Jan 2026 |
Abstract
Soft robots can revolutionize several applications with high demands on dexterity and safety. When operating these systems, real-time estimation and control require fast and accurate models. However, prediction with first-principles (FP) models is slow, and learned black-box models have poor generalizability. Physics-informed machine learning offers excellent advantages here, but it is currently limited to simple, often simulated systems without considering changes after training. We propose physics-informed neural networks (PINNs) for articulated soft robots (ASRs) with a focus on data efficiency. The amount of expensive real-world training data is reduced to a minimum - one dataset in one system domain. Two hours of data in different domains are used for a comparison against two gold-standard approaches: In contrast to a recurrent neural network, the PINN provides a high generalizability. The prediction speed of an accurate FP model is exceeded with the PINN by up to a factor of 467 at slightly reduced accuracy. This enables nonlinear model predictive control (MPC) of a pneumatic ASR. Accurate position tracking with the MPC running at 47 Hz is achieved in six dynamic experiments.
Keywords
- and learning for soft robots, control, model learning for control, Modeling, optimization and optimal control, physics-informed machine learning, Modeling, control, and learning for soft robots
ASJC Scopus subject areas
- Engineering(all)
- Control and Systems Engineering
- Computer Science(all)
- Computer Science Applications
- Engineering(all)
- Electrical and Electronic Engineering
Cite this
- Standard
- Harvard
- Apa
- Vancouver
- BibTeX
- RIS
In: IEEE Transactions on Robotics, Vol. 42, 16.01.2026, p. 619 - 636.
Research output: Contribution to journal › Article › Research › peer review
}
TY - JOUR
T1 - Generalizable and Fast Surrogates
T2 - Model Predictive Control of Articulated Soft Robots using Physics-Informed Neural Networks
AU - Habich, Tim Lukas
AU - Mohammad, Aran
AU - Ehlers, Simon F.G.
AU - Bensch, Martin
AU - Seel, Thomas
AU - Schappler, Moritz
N1 - Publisher Copyright: © 2025 IEEE. All rights reserved.
PY - 2026/1/16
Y1 - 2026/1/16
N2 - Soft robots can revolutionize several applications with high demands on dexterity and safety. When operating these systems, real-time estimation and control require fast and accurate models. However, prediction with first-principles (FP) models is slow, and learned black-box models have poor generalizability. Physics-informed machine learning offers excellent advantages here, but it is currently limited to simple, often simulated systems without considering changes after training. We propose physics-informed neural networks (PINNs) for articulated soft robots (ASRs) with a focus on data efficiency. The amount of expensive real-world training data is reduced to a minimum - one dataset in one system domain. Two hours of data in different domains are used for a comparison against two gold-standard approaches: In contrast to a recurrent neural network, the PINN provides a high generalizability. The prediction speed of an accurate FP model is exceeded with the PINN by up to a factor of 467 at slightly reduced accuracy. This enables nonlinear model predictive control (MPC) of a pneumatic ASR. Accurate position tracking with the MPC running at 47 Hz is achieved in six dynamic experiments.
AB - Soft robots can revolutionize several applications with high demands on dexterity and safety. When operating these systems, real-time estimation and control require fast and accurate models. However, prediction with first-principles (FP) models is slow, and learned black-box models have poor generalizability. Physics-informed machine learning offers excellent advantages here, but it is currently limited to simple, often simulated systems without considering changes after training. We propose physics-informed neural networks (PINNs) for articulated soft robots (ASRs) with a focus on data efficiency. The amount of expensive real-world training data is reduced to a minimum - one dataset in one system domain. Two hours of data in different domains are used for a comparison against two gold-standard approaches: In contrast to a recurrent neural network, the PINN provides a high generalizability. The prediction speed of an accurate FP model is exceeded with the PINN by up to a factor of 467 at slightly reduced accuracy. This enables nonlinear model predictive control (MPC) of a pneumatic ASR. Accurate position tracking with the MPC running at 47 Hz is achieved in six dynamic experiments.
KW - and learning for soft robots
KW - control
KW - model learning for control
KW - Modeling
KW - optimization and optimal control
KW - physics-informed machine learning
KW - Modeling, control, and learning for soft robots
UR - http://www.scopus.com/inward/record.url?scp=105021529003&partnerID=8YFLogxK
U2 - 10.1109/TRO.2025.3631818
DO - 10.1109/TRO.2025.3631818
M3 - Article
AN - SCOPUS:105021529003
VL - 42
SP - 619
EP - 636
JO - IEEE Transactions on Robotics
JF - IEEE Transactions on Robotics
SN - 1552-3098
ER -