Interest in soft robotics has increased in recent years due to their potential in a myriad of applications. A wide variety of soft robots has emerged, including bio-inspired robotic swimmers such as jellyfish, rays, and robotic fish. However, the highly nonlinear fluid-structure interactions pose considerable challenges in the analysis, modeling, and feedback control of these soft robotic swimmers. In particular, developing models that are of high fidelity but are also amenable to control for such robots remains an open problem. In this work, we pro- pose a data-driven approach that exploits Koopman operators to obtain a linear representation of the soft swimmer dynamics. Specifically, two methodologies are explored for obtaining the basis functions of the the operator, one based on data-based derivatives estimated using high-gain observers, and the other based on the dynamics structure of a tail-actuated rigid-body robotic fish. The resulting approximate finite-dimensional operators are trained and evaluated using data from high-fidelity CFD simulations that incorporate fluid-structure interactions. Validation results demonstrate that, while both methods are promising in producing control-oriented models, the approach based on derivative estimates shows higher accuracy in state prediction.
more »
« less
Model Predictive Control-Based Path-Following for Tail-Actuated Robotic Fish
There has been an increasing interest in the use of autonomous underwater robots to monitor freshwater and marine environments. In particular, robots that propel and maneuver themselves like fish, often known as robotic fish, have emerged as mobile sensing platforms for aquatic environments. Highly nonlinear and often under-actuated dynamics of robotic fish present significant challenges in control of these robots. In this work, we propose a nonlinear model predictive control (NMPC) approach to path-following of a tail-actuated robotic fish that accommodates the nonlinear dynamics and actuation constraints while minimizing the control effort. Considering the cyclic nature of tail actuation, the control design is based on an averaged dynamic model, where the hydrodynamic force generated by tail beating is captured using Lighthill's large-amplitude elongated-body theory. A computationally efficient approach is developed to identify the model parameters based on the measured swimming and turning data for the robot. With the tail beat frequency fixed, the bias and amplitude of the tail oscillation are treated as physical variables to be manipulated, which are related to the control inputs via a nonlinear map. A control projection method is introduced to accommodate the sector-shaped constraints of the control inputs while minimizing the optimization complexity in solving the NMPC problem. Both simulation and experimental results support the efficacy of the proposed approach. In particular, the advantages of the control projection method are shown via comparison with alternative approaches.
more »
« less
- NSF-PAR ID:
- 10125711
- Date Published:
- Journal Name:
- Journal of Dynamic Systems, Measurement, and Control
- Volume:
- 141
- Issue:
- 7
- ISSN:
- 0022-0434
- Format(s):
- Medium: X
- Sponsoring Org:
- National Science Foundation
More Like this
-
-
null (Ed.)This paper presents a generalizable methodology for data-driven identification of nonlinear dynamics that bounds the model error in terms of the prediction horizon and the magnitude of the derivatives of the system states. Using higher order derivatives of general nonlinear dynamics that need not be known, we construct a Koopman operator-based linear representation and utilize Taylor series accuracy analysis to derive an error bound. The resulting error formula is used to choose the order of derivatives in the basis functions and obtain a data-driven Koopman model using a closed-form expression that can be computed in real time. Using the inverted pendulum system, we illustrate the robustness of the error bounds given noisy measurements of unknown dynamics, where the derivatives are estimated numerically. When combined with control, the Koopman representation of the nonlinear system has marginally better performance than competing nonlinear modeling methods, such as SINDy and NARX. In addition, as a linear model, the Koopman approach lends itself readily to efficient control design tools, such as LQR, whereas the other modeling approaches require nonlinear control methods. The efficacy of the approach is further demonstrated with simulation and experimental results on the control of a tail-actuated robotic fish. Experimental results show that the proposed data-driven control approach outperforms a tuned PID (Proportional Integral Derivative) controller and that updating the data-driven model online significantly improves performance in the presence of unmodeled fluid disturbance. This paper is complemented with a video: https://youtu.be/9 wx0tdDta0.more » « less
-
This paper presents a data-driven methodology for linear embedding of nonlinear systems. Utilizing structural knowledge of general nonlinear dynamics, the authors exploit the Koopman operator to develop a systematic, data-driven approach for constructing a linear representation in terms of higher order derivatives of the underlying nonlinear dynamics. With the linear representation, the nonlinear system is then controlled with an LQR feedback policy, the gains of which need to be calculated only once. As a result, the approach enables fast control synthesis. We demonstrate the efficacy of the approach with simulations and experimental results on the modeling and control of a tail-actuated robotic fish and show that the proposed policy is comparable to backstepping control. To the best of our knowledge, this is the first experimental validation of Koopman-based LQR control.more » « less
-
null (Ed.)Thanks to their flexibility, soft robotic devices offer critical advantages over rigid robots, allowing adaptation to uncertainties in the environment. As such, soft robots enable various intriguing applications, including human-safe interaction devices, soft active rehabilitation devices, and soft grippers for pick-and-place tasks in industrial environments. In most cases, soft robots use pneumatic actuation to inflate the channels in a compliant material to obtain the movement of the structure. However, due to their flexibility and nonlinear behavior, as well as the compressibility of air, controlled movements of the soft robotic structure are difficult to attain. Obtaining physically-based mathematical models, which would enable the development of suitable control approaches for soft robots, constitutes thus a critical challenge in the field. The aim of this work is, therefore, to predict the movement of a pneumatic soft robot by using a data-driven approach based on the Koopman operator framework. The Koopman operator allows simplifying a nonlinear system by“lifting” its dynamics into a higher dimensional space, where its behavior can be accurately approximated by a linear model, thus allowing a significant reduction of the complexity of the design of the resulting controllers.more » « less
-
more » « less
Our goal is to learn control policies for robots that provably generalize well to novel environments given a dataset of example environments. The key technical idea behind our approach is to leverage tools from generalization theory in machine learning by exploiting a precise analogy (which we present in the form of a reduction) between generalization of control policies to novel environments and generalization of hypotheses in the supervised learning setting. In particular, we utilize the probably approximately correct (PAC)-Bayes framework, which allows us to obtain upper bounds that hold with high probability on the expected cost of (stochastic) control policies across novel environments. We propose policy learning algorithms that explicitly seek to minimize this upper bound. The corresponding optimization problem can be solved using convex optimization (relative entropy programming in particular) in the setting where we are optimizing over a finite policy space. In the more general setting of continuously parameterized policies (e.g., neural network policies), we minimize this upper bound using stochastic gradient descent. We present simulated results of our approach applied to learning (1) reactive obstacle avoidance policies and (2) neural network-based grasping policies. We also present hardware results for the Parrot Swing drone navigating through different obstacle environments. Our examples demonstrate the potential of our approach to provide strong generalization guarantees for robotic systems with continuous state and action spaces, complicated (e.g., nonlinear) dynamics, rich sensory inputs (e.g., depth images), and neural network-based policies.
- Free Publicly Accessible Full Text
-
Accepted Manuscript1.0
- Journal Article:
- https://doi.org/10.1115/1.4043152
