Modelling and Control of Soft Robotic Manipulators

Workshop on Modelling and Control of Soft Robotic Manipulators
 

MAIN CONTENT

  1. MOTIVATION AND OBJECTIVES

The Workshop on Modelling and Control of Soft Robotic Manipulators will be organized on April 24th, at the first IEEE-RAS International Conference on Soft Robotics – RoboSoft 2018, that will be held in Livorno,Italy.

In the past decade, a novel sub-domain of continuum manipulators, referred to as soft robotic manipulators, has been rapidly growing. They were inspired from boneless biological organisms such as tentacles, trunks or tongues which are able to exploit the mechanically intelligent arrangement of just their muscles to exhibit dexterous advanced manipulation capabilities in cluttered environments. This has been translated into new range of continuum manipulators made up of soft materials, such as silicone, due to their ability to undergo a large deformation under normal operation. The underlying idea is to use principles of embodied intelligence and morphological computation to exploit the soft material properties to enable machines with properties such as inherent compliance, variable stiffness, and highly dexterous motion in unstructured environment. The resulting systems have the ability to simplify a wide range of well-known complex tasks. Furthermore, the deformability of the soft material offers compliance which facilitates safe human-robot interaction in comparison to their rigid counterparts. These desirable characteristics are the fundamental reasons behind their rapidly increasing demand in industrial, surgical,and assistive applications.

However, the long-term success for the practical application of these systems is dependent up on the development of real-time kinematic and/or dynamic models and controllers that facilitate fast, reliable, accurate, and energy-efficient control. This is non-trivial because: (i) unlike rigid manipulators, the movement of which can be specified by three translations and three rotations, elastic deformation of soft robotic manipulators results in virtually infinite degrees-of- freedom (DoF) motions, (bending, extension, contraction, torsion, buckling, etc.) (ii) the material properties exhibit non-linear characteristics such as compliance and hysteresis that restricts high-frequency control (iii) the wide range of design and actuation techniques which makes each of these robots have unique properties.

This workshop aims to provide an insight into the various methodologies for modeling and control of soft robotic manipulators as a guideline for future applications in the soft robotics field. The final goal is to bring together researchers from both modeling and control in order to explore how to maximize the progress evaluating the advantages of model-free and model based controllers.

TUTORIAL SESSION : SOFA PLATFORM

Numerical mechanics can provide some key answers for the design, the simulation and the control of soft robots. The tools developed by the  DEFROST team (Inria, University of Lille, CNRS, France) on the open-source framework SOFA allow to model, simulate and control soft robots in real-time using the Finite Element Method. The approach is very generic and applicable to many existing soft robots and could become an important platform for the soft robotics community. 

The special tutorial session presents these tools and some recent research results. During this tutorial, the main features of the platform will be illustrated: creation of the soft robot mechanical model using FEM, simulation with contact modeling, inverse model, and control methods. Organizers will bring 10 soft robots that could be connected through USB so that attendees can directly practice during the workshop on their own laptop or with their neighbors. Laptops should have at least 4GB of free space to be able to use the simulation software.

The tutorial will be presented by Olivier Goury with the help of people from DEFROST team that will be in the room for helping you if you have questions during the tutorial. 

The presentation will be structured as followed:

| Context and motivations
| Step by step modeling of the robot 
| More advances features: Contact, Inverse modeling, Closed loop…

 

SPEAKERS

  • Robert Katzschmann [MIT]
Title: From Open Loop to Model-based Control
Abstract: Soft and compliant robotic designs are inherently well-suited for manipulating objects and interacting with an environment. Despite the emergence of many soft-bodied robotic systems, feedback control has remained an open challenge. This is largely due to the intrinsic difficulties in designing controllers for systems with infinite dimensions. In this talk, I will present a range of approaches for controlling soft robots. I will cover open-loop control for underwater locomotion, closed-loop control for soft manipulator arms performing grasps and entering confined spaces, and model-based control for actuated soft surfaces and soft manipulator arms. With the focus on model-based control, the first example demonstrates dynamic object manipulation with an actuated soft surface. The model-based control of this system is realized through proposing an implicit minimal parameter model capturing the dynamics of the soft contact between a rigid puck and a highly deformable surface. The second examples is the dynamic control of a pneumatically-actuated multi-segment soft manipulator arm. The modeling of that system is enabled by an alternative formulation of the dynamics connecting the soft robot’s behavior with the one of a rigid-bodied robot with elasticities in the joints. Based on this connection, two control architectures are introduced, with the aim of either achieving accurate curvature control or accomplishing Cartesian regulation of the end-effector’s impedance. The first architecture is a curvature controller that accounts for the natural softness of the system. The second architecture is a Cartesian controller that adapts the impedance of the end effector for interactions with an environment. 
 
  • Caleb Rucker [University of Tennesse]
Title: Elastic Rod Models for Real-Time Simulation and Control of Soft and Continuum Robots

Abstract: This talk will review and highlight a Cosserat-rod modeling framework applied to concentric-tube robots, tendon-driven continuum robots, fluid-driven soft robots, and parallel continuum robots. In this framework, efficient algorithms for solving the model can enable real-time simulation and teleoperation. Model linearization provides four matrices: the Jacobian, the output compliance matrix, the input stiffness matrix, and the force reflectivity matrix, all of which can be used to characterize and visualize aspects of robot performance via ellipsoids. We will also show how the same underlying framework can be used to formulate and solve several different problem statements: forward kinematics with known loads, inverse kinematics with known loads, deflection-based force sensing, and actuation-based force sensing. This paves the way for future work in advanced control of soft robot interactions via impedance control and hybrid force/position control.

 

  • Federico Renda [Khalifa University]
Title: A Geometric Theory of Hybrid (Soft-Rigid) Multi-body Systems
Abstract: The continuum nature of the soft robots’ structure has driven the research on soft robots modeling on clearly diverging paths with respect to the well-established rigid robots counterparts. Nevertheless, a unified theory able to connect these two realms would be of tremendous benefit for the soft robotics community to exploit the powerful tools already available for the rigid links ancestors. In this presentation, a plausible candidate for such unifying theory is presented, where the geometric theory of robotics developed since Brockett’s original work on the subject is generalized to hybrid (soft-rigid) systems thanks to a discrete Cosserat approach of the soft-body dynamics. This generalized geometric theory includes multidimensional joints and flexible beams with different kinematic constrains under the same framework. In fact, under the lens of this theory, a rigid joint reduces to a special case of immaterial flexible beam with unitary length, restoring the broken ties between rigid and soft robotics. To complete the picture, a generalization to hybrid systems of the recursive Newton-Euler algorithms, traditionally devoted to rigid multi-body dynamics, is also presented. These extended algorithms are able to solve inverse, forward or mixed hybrid multi-body dynamic problems with linear O(N) complexity with respect to the number of bodies in the system.

 

  • Helmut Hauser [University of Bristol]
Title: To Control, or not to Control, that is the Question.
Abstract: Despite the success of Soft Robotics as a research field, it is still struggling with very basic problems like how to model and control soft machines. The reason is that their complex and nonlinear dynamics, which are often perceived as beneficial and exploited in clever mechanisms, are also hard to model and control. While on the design side Soft Robotics is finally breaking with classical approaches by using materials and actuation systems that go beyond rigid bodies and electric motors, we still think along conventional lines when it comes to controlling them. We believe that we are missing an opportunity by concentrating only on adapting and extending classical control approaches. In order to take full advantage of this paradigm shift that Soft Robotics is offering, we also have to embrace novel ways to think about control. One possible starting point is provided by morphological computation and embodiment. These concepts suggest that instead of modelling and controlling every aspect of the system, we should leave room for the emergence of interesting behaviour and allow the robot to interact with its environment. Instead of "how to control" we have to ask rather "what to control." This leads to novel concepts like morphological communication and control, morphosis, orchestration, behaviour through growing, and design by stimulation and for emergence. 
 
 
  • Rochdi Merzouki [University of Lille]
Title: Towards Cooperative Robotics: Manufacturing and Healthcare
Abstract: Today, different types of intelligent mechatronic systems are in use or under development for different purposes. Among them, we can cite autonomous robots. One important goal is to use these robots at home or in hospital. For this reason, new concept of robots are designed, often inspired from nature and built with soft materials and smart sensors. The performance of these robots should assist elder people in eating, drinking, compliant grasping tasks etc or for surgery and therapeutically purposes. For such application, nowadays, it is needed to develop a class of robot arms with medium performances, in terms of speed and accuracy, comparing to professional arms with rigid bodies and toys arms, but with reasonable cost-effective. Such mechatronics robots should interact smoothly with the patient during the various interventions. For this purpose, development of lightweight robot arm should integrate for their design: a mechanical structure made by rapid manufacturing, a low cost and efficient multi-domain actuators with wire-potentiometer sensors, bionic and compliant gripper, and mainly an accurate control.

 

  • Thomas George Thuruthel [Scuola Superiore Sant’Anna]
Title: Model-free Control strategies for soft robotic manipulators 
Abstract: This talk overviews few machine-learning based control strategies for soft robotic manipulators. Due to their highly nonlinear characteristics, intrinsic stochasticity and design variability analytically modeling these systems become very difficult to develop or are not easily transferable. In this scenario, models developed from empirical observations becomes a viable option. The presentation will cover approaches for easily developing closed loop inverse kinematic controllers, open loop dynamic controllers and closed loop task specific dynamic controllers. A brief comparison of their advantages and limitations over analytical models will be done. Experimental results showcasing the performance of each of these controllers and a deep analysis into their accuracy, robustness and stability with the aim to advocate their wider application will be presented.

 

  • Yasmin Ansari [Scuola Superiore Sant’Anna]
Title: Enabling High-precision Positioning Capabilities in Soft Robotic Manipulators: Challenges and Methods
Abstract: Continuum manipulators have found an increasing demand in industrial and healthcare environments due to their appealing advantages of compliance, dexterity, and variable stiffness. One of the fundamental uses of continuum manipulators is for positioning in 3D space. However, their use in practical applications is non-trivial due to energy losses from internal frictional forces which cause positional inaccuracies. Internal friction is a resistive force that arises in reaction to two interacting surfaces that undergo relative motion. In the given context, examples of interacting surfaces include, but are not limited to, a pneumatic membrane with external fiber-reinforcements, tendons routed along the backbone, etc. Depending upon the speed of the relative motion, two nonlinear frictional regimes arise: (i) hysteresis - where the speed of the relative motion is close to zero and the system undergoes elastic deformation. It is primarily characterized by a time-independent multi-valued function; (ii) frictional-lag and stribeck effect - where the speed of the relative motion a function of input velocity and is characterized by a time-dependent single-valued function. For position controllers that function under the assumption of quasi-static behavior, the speed of relative motion is close to zero. As a result, such input-output mappings are predominantly affected by the nonlinear frictional effects of hysteresis which must be compensated in order to achieve high performance. This talk presents the tools and mechanisms necessary to develop model-free algorithms for high-precision position control.
 
  • Marc Killpack [Brigham Young University]
Title: Past success and future directions for modeling and control of large-scale soft robot manipulators

Abstract: Over the past 3.5 years, our research group has worked in collaboration with the startup company Pneubotics to develop control and modelling methods for large-scale, soft, fabric-based, pneumatic, robot manipulators. In the first part of my presentation, I will share our control and modeling methods for successful configuration and end effector control, variable stiffness control, adaptive control for unknown payloads and failure mitigation. In large part, these methods have focused on simplified gas dynamics and soft robot models that were tractable for model predictive control. In the 2nd part of my talk, I will share future directions that we expect to be important in continuing to improve the performance of our soft robots for real-world tasks. These areas of research include design for soft robot control, using parallel computation to solve optimal control problems for large numbers of degrees of freedom, and using machine learning to improve our model prediction capability for control. 
 
  • Kaspar Althoefer [Queen Mary University of London]
Title: Modelling and control using lumped parameter models for soft material robots

Abstract: Soft material robot arms are highly flexible structures and achieve motion due to a deformation that is distributed along their bodies. They contain no discrete joints whose position can be easily measured and no rigid links that can be easily modelled. For that reason, traditional techniques have only limited use for modelling and controlling soft material robot arms, especially in scenarios where the robot is in contact with the environment. Soft material modelling can be achieved using general methods such as the Finite Element Method (FEM). The FEM approach is to divide the structure to be modelled into very small pieces, model the deformations for each of these pieces and integrate the results, to obtain the deformation of the overall structure. There have been significant advancements in FEM for real-time control; still high update rates remain difficult to achieve. Another approach to soft material robot modelling can be achieved by employing simplified lump-parameter models that are based on the physical descriptions of the behaviour of the used material. The models are tailored to the particular robot structure modelled. Only factors that impact on the robot motion are taken into account and only the deformations that are important are calculated; this allows us to create models that are lightweight and can be calculated within a real-time control loop. Despite the considerable simplifications introduced, the proposed approach is capable of computing the pose of the robot arm fairly accurately in different scenarios. The performance of the proposed models is studied especially considering dynamics and physical interactions with the environment.

 

  • Cosimo Della Santina [Centro E. Piaggio - University of Pisa]
Title: Controlling elastic robots: from articulated to continuous soft robots.

Abstract: Animals move very differently from rigid robots, performing tasks efficiently, and interacting compliantly with the external world. With the aim of getting closer to animals’ performance, elastic elements are introduced in the mechanical structure of soft robots. On the basis of their main inspiration, soft robots can be divided in two classes. Articulated soft robots are inspired by the vertebrates’ muscle-skeletal system and include elastic elements in a rigid body. Continuous soft robots take instead inspiration from the invertebrate body, and their compliance is distributed in the whole structure. When it comes to control, however, it turns out that achieving performance is not at all easy. This is intuitive for such indices as accuracy, which is the reason why industrial robots have traditionally been built for maximum rigidity. Thanks to their more conventional nature, articulated soft robots have been shown to be much simpler to control. On the other hand, the maximum level of performance is promised by the continuous kind. However, recent progresses in modelling continuous soft robots are pointing out how continuous and articulated soft robots are less fundamentally different than previously expected. In this talk, I will present preliminary results in applying dynamic feedback control to continuous soft robots. I will then focus on recent achievements in controlling articulated soft robots that could inform novel techniques for continuous soft robots. These include a recently emerged fundamental limitation in feedback control, i.e. that good performance implies de facto to stiff the robot, and a new method for producing scalable controllers taking full advantage of the intelligence embodied in the soft robot dynamics.

PROGRAM





Time

Agenda

Speaker

Institution

09:00 – 9:10

Opening Talk

Organizer

Scuola Superiore Sant’Anna

9:10 – 9:40

To Control, or not to Control, that is the Question

Helmut Hauser

University of Bristol

   9:40 – 10:10

Recipes for Controlling Soft Robots – From Open Loop to Model-based Control

Robert  Katzschmann

MIT

10:10 -  10:30

Model-free Control strategies for soft robotic manipulators ​

Thomas George Thuruthel

Scuola Superiore Sant’Anna

10:30 - 11:00

Coffee Break

11:00 - 11:30

Elastic Rod Models for Real-Time Simulation and Control of Soft and Continuum Robots

Caleb Rucker

University of Tennesse

11:30 – 12:00

A Geometric Theory of Hybrid (Soft-Rigid) Multi-body Systems

Federico Renda

Khalifa University

 12-00 – 12:30

Past success and future directions for modeling and control of large-scale soft robot manipulators

Marc Killpack

Brigham Young University

 12-30 – 13:00

Modelling and control using lumped parameter models  for soft material robots

Kaspar Althoefer

Queen Mary University of London

13:00 – 14:00

Free time for lunch

 14-00 – 14:30

Towards Cooperative Robotics:

Manufacturing and Healthcare

Rochdi Merzouki

CRIStAL, University of Lille

14:30 -  14:50

Controlling elastic robots: from articulated to continuous soft robots ​

Cosimo Della Santina

Centro E. Piaggio – University of Pisa

14:50 - 15:20

SOFA-Mechanics, FEM

Contact and inverse modeling

Olivier Goury,

Eulalie Coevoet,

INRIA, Lille1 University

15:20 - 16:00

SOFA- Closed loop control

Thor Bieze,

Alexandre Kruszewski,

INRIA, Lille1 University

16:00- 16:30

 

Coffee Break

 

16:30 - 17:00

SOFA- Practice Session

INRIA team

INRIA, Lille1 University

17:00- 17:20

Enabling High-precision Positioning Capabilities in Soft Robotic Manipulators: Challenges and Methods

Yasmin Ansari

Scuola Superiore Sant’Anna

17:20 - 17:30

Presentation of selected abstracts (I)

Maria Pozzi (University of Siena) -  Efficient FEM-Based Simulation of Soft Fingers

17:30 - 17:40

Presentation of selected abstracts (II)

Haider Abidi (Scuola Superiore Sant’Anna) - Design inspirations from the various Models of Soft robots

17:40 - 17:50

Presentation of selected abstracts (III)

Stanislao Grazioso (University of Naples Federico II) - A new finite element solver for soft robots dynamics

17:50 - 18:00

Closing Remarks

Venue

 

Sala Fattori 
Grand Hotel Palazzo
Viale Italia
Livorno
Italy

 

Authors Area

Abstract Submission 

 

Important Dates

• Call for Contributions:
March 30, 2018 

• Notification of acceptance:
April 6, 2018

 

Organizers