Soft Robot Control With a Learned Differentiable Model

Bern, James M.; Schnider, Yannick; Banzet, Pol; Kumar, Nitish; Coros, Stelian

doi:10.1109/robosoft48309.2020.9116011

Cited by 75 publications

(40 citation statements)

References 34 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Their work uses sensitivity analysis to obtain gradients from the dynamic equations but conducts system identification through trial and error. Finally, [15] proposes to control a soft tendon with a learned differentiable model. Compared to our pipeline, [15] uses a neural network model and assumes the motion is quasi-static, while our pipeline removes the quasistatic assumption and leverages an analytic dynamic model.…”

Section: Related Workmentioning

confidence: 99%

“…Our core idea is to embed a differentiable simulator into a pipeline that alternates between simulated and real experiments. With gradient information readily available from a differentiable simulator, previous papers have demonstrated promising results in various soft-robot applications, including system identification and controller design [11]- [15]. However, results from existing differentiable simulators are primarily focused on simulated robots, and demonstrations on real underwater soft robots have yet to be seen.…”

Section: Introductionmentioning

confidence: 99%

See 1 more Smart Citation

Underwater Soft Robot Modeling and Control With Differentiable Simulation

Hughes

Wah

et al. 2021

IEEE Robot. Autom. Lett.

View full text Add to dashboard Cite

Underwater soft robots are challenging to model and control because of their high degrees of freedom and their intricate coupling with water. In this paper, we present a method that leverages the recent development in differentiable simulation coupled with a differentiable, analytical hydrodynamic model to assist with the modeling and control of an underwater soft robot. We apply this method to Starfish, a customized soft robot design that is easy to fabricate and intuitive to manipulate. Our method starts with data obtained from the real robot and alternates between simulation and experiments. Specifically, the simulation step uses gradients from a differentiable simulator to run system identification and trajectory optimization, and the experiment step executes the optimized trajectory on the robot to collect new data to be fed into simulation. Our demonstration on Starfish shows that proper usage of gradients from a differentiable simulator not only narrows down its simulation-to-reality gap but also improves the performance of an open-loop controller in real experiments.

show abstract

Section: Related Workmentioning

confidence: 99%

Section: Introductionmentioning

confidence: 99%

Underwater Soft Robot Modeling and Control With Differentiable Simulation

Hughes

Wah

et al. 2021

IEEE Robot. Autom. Lett.

View full text Add to dashboard Cite

show abstract

“…Data-driven machine learning methods may provide alternative solutions for the design and control of soft robots in the lack of existing analytical or numerical models that describe their underlying kinematics, dynamics, and functions (Chin et al, 2020). One common approach is to learn these models by gathering data from robot experiments and training a neural network (NN) architecture (Bern et al, 2020;Hyatt et al, 2019;Thuruthel et al, 2019). However, the need for data efficiency, i.e., the ability to learn from only a few experimental trials, presents a core challenge for such methods (Chatzilygeroudis et al, 2019).…”

Section: Introductionmentioning

confidence: 99%

Task space adaptation via the learning of gait controllers of magnetic soft millirobots

Demır

Çulha

Karacakol

et al. 2021

The International Journal of Robotics Research

View full text Add to dashboard Cite

Untethered small-scale soft robots have promising applications in minimally invasive surgery, targeted drug delivery, and bioengineering applications as they can directly and non-invasively access confined and hard-to-reach spaces in the human body. For such potential biomedical applications, the adaptivity of the robot control is essential to ensure the continuity of the operations, as task environment conditions show dynamic variations that can alter the robot’s motion and task performance. The applicability of the conventional modeling and control methods is further limited for soft robots at the small-scale owing to their kinematics with virtually infinite degrees of freedom, inherent stochastic variability during fabrication, and changing dynamics during real-world interactions. To address the controller adaptation challenge to dynamically changing task environments, we propose using a probabilistic learning approach for a millimeter-scale magnetic walking soft robot using Bayesian optimization (BO) and Gaussian processes (GPs). Our approach provides a data-efficient learning scheme by finding the gait controller parameters while optimizing the stride length of the walking soft millirobot using a small number of physical experiments. To demonstrate the controller adaptation, we test the walking gait of the robot in task environments with different surface adhesion and roughness, and medium viscosity, which aims to represent the possible conditions for future robotic tasks inside the human body. We further utilize the transfer of the learned GP parameters among different task spaces and robots and compare their efficacy on the improvement of data-efficient controller learning.

show abstract

“…We can think of a simulator as a highly complex function, taking forces at a discrete set of locations as input, returning a deformed configuration as output. There are two incentives to replace this function at least partially with a neural network: (1) to augment the simulation model where the sim-to-real gap is high and more accurate models are unknown (e.g., for frictional contact [26]), and (2) to reduce the time complexity of simulations to either speed up simulation-driven design and control [27, 28 •], or to enable the interactive exploration of soft robot designs [29 •]. There is also work on completely replacing a simulator with a neural network [30], learned from either real or simulated data. While providing an exciting avenue for future research, a complete replacement of the simulator comes at the cost of longer training and requiring more data.…”

Section: Differentiable Simulation and Learningmentioning

confidence: 99%

Design and Control of Soft Robots Using Differentiable Simulation

2021

View full text Add to dashboard Cite

Purpose of Review We discuss the use of differentiable simulation for computational problems in soft robotics. This includes characterizing the mechanical behavior of soft robots, optimally controlling embedded soft actuators or active materials, and estimating the robot's state from readings of embedded sensors. Moreover, we discuss how design optimization can help to optimally place soft actuators and sensors. Recent Findings We expatiate on the adoption of simulation and optimization tools in the process of designing and controlling soft robots. We include a discussion of rigid-flexible systems and the use of differentiable simulation in combination with machine learning. Summary We review the state of the art in the computational modeling of soft robots and provide a summary of the required mathematical tools. We also review several open questions where computation could help to move the field forward, and discuss the role of differentiable simulation in managing the ever-growing design complexity of next-generation soft robots. Keywords Differentiable simulation • Computational design and control • State estimation • Sensor and actuator design This article is part of the Topical Collection on Soft Robotics

show abstract

Soft Robot Control With a Learned Differentiable Model

Cited by 75 publications

References 34 publications

Underwater Soft Robot Modeling and Control With Differentiable Simulation

Underwater Soft Robot Modeling and Control With Differentiable Simulation

Task space adaptation via the learning of gait controllers of magnetic soft millirobots

Design and Control of Soft Robots Using Differentiable Simulation

Contact Info

Product

Resources

About