Learning-Based Nonlinear Model Predictive Control of Articulated Soft Robots Using Recurrent Neural Networks DOI

Hendrik Schäfke,

Tim-Lukas Habich,

Christian Muhmann

et al.

IEEE Robotics and Automation Letters, Journal Year: 2024, Volume and Issue: 9(12), P. 11609 - 11616

Published: Nov. 11, 2024

Soft robots pose difficulties in terms of control, requiring novel strategies to effectively manipulate their compliant structures. Model-based approaches face challenges due the high dimensionality and nonlinearities such as hysteresis effects. In contrast, learning-based provide nonlinear models different soft based only on measured data. this paper, recurrent neural networks (RNNs) predict behavior an articulated robot (ASR) with five degrees freedom (DoF). RNNs gated units (GRUs) are compared more commonly used long short-term memory (LSTM) show better accuracy. The recurrence enables capture effects that inherent viscoelasticity or friction but cannot be captured by simple feedforward networks. data-driven model is within a predictive control (NMPC), whereby correct handling RNN's hidden states focused. A training approach presented allows values utilized each cycle. This accurate predictions short horizons sensor data, which crucial for closed-loop NMPC. proposed NMPC trajectory tracking average error 1.2deg experiments pneumatic five-DoF ASR.

Language: Английский

Motion Control for Continuum Robots: A Mini Review for Model-Free and Hybrid-Model Control DOI
Zhimin Du, Laihao Yang,

Yu Sun

et al.

Lecture notes in computer science, Journal Year: 2025, Volume and Issue: unknown, P. 372 - 391

Published: Jan. 1, 2025

Language: Английский

Citations

0
A Holistic Indirect Contact Identification Method for Soft Robot Proprioception DOI
Shuoqi Wang, Keng-Yu Lin, Xiangru Xu

et al.

Soft Robotics, Journal Year: 2025, Volume and Issue: unknown

Published: Feb. 24, 2025

Soft robots hold great promise but are notoriously difficult to control due their compliance and back-drivability. In order implement useful controllers, improved methods of perceiving robot pose (position orientation the entire body) in free perturbed states needed. this work, we present a holistic approach perception bending with external contact, using multiple soft strain sensors on (not collocated point contact). By comparing deviation these from value an unperturbed pose, able perceive mode magnitude deformation thereby estimate resulting actuator. We develop sample 2 degree-of-freedom finger two sensors, characterize sensor response front, lateral, twist perturbation. data-driven model free-bending deformation, impose our perturbation method, demonstrate ability single-finger two-finger gripper. Our contact identification method provides generalizable needed for robots.

Language: Английский

Citations

0
Soft Materials and Devices Enabling Sensorimotor Functions in Soft Robots DOI

Jiangtao Su,

Ke He, Yanzhen Li

et al.

Chemical Reviews, Journal Year: 2025, Volume and Issue: unknown

Published: March 31, 2025

Sensorimotor functions, the seamless integration of sensing, decision-making, and actuation, are fundamental for robots to interact with their environments. Inspired by biological systems, incorporation soft materials devices into robotics holds significant promise enhancing these functions. However, current systems often lack autonomy intelligence observed in nature due limited sensorimotor integration, particularly flexible sensing actuation. As field progresses toward soft, flexible, stretchable materials, developing such becomes increasingly critical advanced robotics. Despite rapid advancements individually devices, combined applications enable capabilities emerging. This review addresses this emerging providing a comprehensive overview that functions robots. We delve latest development technologies, actuation mechanism, structural designs, fabrication techniques. Additionally, we explore strategies control, artificial (AI), practical application across various domains as healthcare, augmented virtual reality, exploration. By drawing parallels aims guide future research robots, ultimately adaptability unstructured

Language: Английский

Citations

0
Learning from Octopuses: Cutting-Edge Developments and Future Directions DOI Creative Commons

Jinjie Duan,

Yuebao Lei, Jie Fang

et al.

Biomimetics, Journal Year: 2025, Volume and Issue: 10(4), P. 224 - 224

Published: April 4, 2025

This paper reviews the research progress of bionic soft robot technology learned from octopuses. The number related papers increased 760 in 2021 to 1170 2024 (Google Scholar query), with a growth rate 53.95% past five years. These studies mainly explore how humans can learn physiological characteristics octopuses for sensor design, actuator development, processor architecture optimization, and intelligent optimization algorithms. tentacle structure nervous system octopus have high flexibility distributed control capabilities, which is an important reference design robots. In terms technology, flexible strain sensors suction cup inspired by achieve accurate environmental perception interaction. Actuator uses muscle fibers movement patterns develop various driving methods, including pneumatic, hydraulic electric systems, greatly improves robot’s motion performance. addition, inspires multi-processor also introduces concept expected functional safety first time safe robots failure or unknown situations. Currently, there are more technologies that draw on octopuses, their application areas constantly expanding. future, further integration artificial intelligence materials science, show greater potential adapting complex environments, human–computer interaction, medical applications.

Language: Английский

Citations

0
Robotic surgery DOI
Gastone Ciuti, Robert J. Webster, Ka‐Wai Kwok

et al.

Nature Reviews Bioengineering, Journal Year: 2025, Volume and Issue: unknown

Published: April 22, 2025

Citations

0
A Hybrid Adaptive Controller for Soft Robot Interchangeability DOI Creative Commons
Zixi Chen, Xuyang Ren, Matteo Bernabei

et al.

IEEE Robotics and Automation Letters, Journal Year: 2023, Volume and Issue: 9(1), P. 875 - 882

Published: Nov. 30, 2023

Soft robots have been leveraged in considerable areas like surgery, rehabilitation, and bionics due to their softness, flexibility, safety. However, it is challenging produce two same soft even with the mold manufacturing process owing complexity of materials. Meanwhile, widespread usage a system requires ability replace inner components without highly affecting performance, which interchangeability. Due necessity this property, hybrid adaptive controller introduced achieve interchangeability from perspective control approaches. This method utilizes an offline-trained recurrent neural network cope nonlinear delayed response robots. Furthermore, online optimizing kinematics applied decrease error caused by above controller. pneumatic different deformation properties but included for validation experiments. In experiments, systems actuation configurations follow desired trajectory errors $\mathbf{3.3\pm 2.9\%}$ notation="LaTeX">$\mathbf{4.3\pm 4.1\%}$ compared working space length, respectively. Such also shows good performance on frequencies velocities. model-based simulation. endows potential wide application, future work may include offline controllers. A weight parameter adjusting strategy be proposed future.

Language: Английский

Citations

6
Learning-Based Nonlinear Model Predictive Control of Articulated Soft Robots Using Recurrent Neural Networks DOI

Hendrik Schäfke,

Tim-Lukas Habich,

Christian Muhmann

et al.

IEEE Robotics and Automation Letters, Journal Year: 2024, Volume and Issue: 9(12), P. 11609 - 11616

Published: Nov. 11, 2024

Soft robots pose difficulties in terms of control, requiring novel strategies to effectively manipulate their compliant structures. Model-based approaches face challenges due the high dimensionality and nonlinearities such as hysteresis effects. In contrast, learning-based provide nonlinear models different soft based only on measured data. this paper, recurrent neural networks (RNNs) predict behavior an articulated robot (ASR) with five degrees freedom (DoF). RNNs gated units (GRUs) are compared more commonly used long short-term memory (LSTM) show better accuracy. The recurrence enables capture effects that inherent viscoelasticity or friction but cannot be captured by simple feedforward networks. data-driven model is within a predictive control (NMPC), whereby correct handling RNN's hidden states focused. A training approach presented allows values utilized each cycle. This accurate predictions short horizons sensor data, which crucial for closed-loop NMPC. proposed NMPC trajectory tracking average error 1.2deg experiments pneumatic five-DoF ASR.

Language: Английский

Citations

0