The workshop will be held on October 5th from 9:00 to 19:00 during IROS 2018 in Madrid, Spain. The workshop will be held in room Amsterdam 2.L.3. The workshop number on Papercept is FrA3 and on the number on the IROS website is FR-WS12.
Several experts will alternate in talks of 30 mins each. The speakers will cover the areas of Articulated Soft Robot Control, Soft Modeling, and Soft-bodied Control. We will ask each speaker to accept questions throughout the talk in order to encourage discussions and the involvement by the audience. We will further encourage the speaker to not only use slide shows, but to make use of the white-boards to introduce key concepts in modeling and control.
For the three poster sessions of the workshop, the poster preentations were selected from a call for extended abstracts. Those who wanted to participate were required to submit an extended abstract together with a poster, see Call for Contributions. An overview of the poster sessions and a link to the papers can be found at the Poster Presentations page.
|Time||Talk||Topic Area: Title|
|09:00 - 09:10||Welcome by Organizers||All topics|
|09:10 - 09:40||Alessandro De Luca||Articulated Soft Robot Control: A review on the control of flexible joint manipulators|
|09:45 - 10:15||Alin Albu-Schäffer||Articulated Soft Robot Control: Nonlinear elastic resonance modes for efficient robot and biological locomotion|
|10:20 - 10:50||Stefano Stramigioli||Articulated Soft Robot Control: Passivity-based approaches for motion and interaction control|
|10:55 - 11:00||1. Poster Teasers|
|11:00 - 11:45||1. Poster Session
(Coffee Break 11:00-11:30)
|11:45 - 12:15||Marco Hutter||Articulated Soft Robot Control: Model-based control of manipulators and walking robots with (high-)compliant series elastic actuators|
|12:20 - 12:50||
Soft-bodied Modeling & Control: Soft robot bodies: a control challenge or a solution for robot control?
|12:55 - 13:25||Federico Renda||Soft-bodied Modeling: Screw Theory as a Unified Approach for Rigid, Soft Articulated and Soft Continuum Robots|
|13:30 - 13:35||2. Poster Teasers|
|13:35 - 14:35||Lunch Break|
|14:35 - 15:20||2. Poster Session cont.|
|15:20 - 15:50||Christian Duriez||Soft-bodied Modeling: Physics based modeling of deformable robots for real-time applications like simulation and control|
|15:55 - 16:25||Elias Cueto||Soft-bodied Modeling: Reduced order modeling of soft robots|
|16:30 - 16:35||3. Poster Teasers|
|16:35 - 17:20||3. Poster Session
(Coffee Break 16:30-17:00)
|17:20 - 17:50||Kohei Nakajima||Soft-bodied Control: Linking an information transfer network and memory in soft body dynamics|
|17:55 - 18:25||Nick Cheney||Soft-bodied Control: The Evolution and Development of Soft Robot Morphologies and Controllers|
|18:55 - 19:00||Closing Remarks||All topics|
Title: A review on the control of flexible joint manipulators
Abstract: I will review the main control results for lightweight compliant manipulators that display non-negligible flexibility concentrated at the joints. Moving from the fully rigid to the flexible case, classical and more recent solutions to set-point regulation and output trajectory tracking tasks will be critically evaluated in terms of model information, complexity of implementation, and performance. The role of (partial/full, linear/nonlinear) state feedback and (model-based/iteratively learned) of torque feedforward is assessed for the cancellation or compensation of dynamic terms and, more in general, for achieving feedback equivalence to a desired reference model, together with error stabilization. In robots with elastic joints, both the fourth-order inverse dynamics and the feedback linearization control can be computed numerically in an efficient way using a modified Newton-Euler scheme. The extension of these results to the case of joints with variable stiffness actuation will be briefly illustrated. Finally, I will discuss some control issues (related to the stable or unstable zero dynamics of the system) that arise when flexibility is distributed along the robot links, as highlighted through a simple, finite-dimensional example.
Title: Nonlinear elastic resonance modes for efficient robot and biological locomotion
Abstract: Controlling motion robustly and at low energetic cost, both from mechanical and computational point of view, certainly constitutes one of the major locomotion challenges in biology and robotics. We attempt to demonstrate that robots can be designed and controlled to walk highly efficient by exploiting resonance body effects, increasing the performance compared to rigid body designs. To do so, however, legged robots need to achieve limit cycle motions of the highly coupled, non-linear body dynamics. This led us to the research of the still not very well understood theory of nonlinear system intrinsic modal oscillation control. I will present current theoretical and experimental results therewith. One of the striking results is that biomechanics, in particular kinematics, visco-elastic and inertial properties of biological limbs are such that coordinated resonant motions of multiple joints intrinsically emerges and is therefore easy to excite and sustain. This can be also achieved by careful design for robotic systems. Moreover, I will present a significant extension of our previous work on compliance control of flexible joint robots, which allow implementing all control features of the DLR-light-weight robots also on highly, elastic, Variable Compliance Robots, such as the DLR humanoid David. Some of the basic robotics control functions we developed for locomotion strikingly resemble neural functionalities and structures. For example, Hebbian lerning, one of the most basic principles of synaptic plasticity, is mathematically equivalent to robotic controllers which adapt to previously unknown resonance properties of the body. Based on the robot control approach, we propose an equivalent neural model involving neural plasticity in the spine and the serotonergic loop in the brain-stem. This hypothesis is supported by numerous experimental evidences from neuroscience.
Title: Passivity-based approaches for motion and interaction control
Abstract: Interaction with the environment means that dynamically, the environment will become part of the system we are controlling meanwhile interacting, but the environment is not known a priory. How can we handle control in such situations ensuring that the interaction will be safe and stable all the times and under a possible distribute architecture of the robot? This may seem a complicated problem, but using concepts of Energy Aware Robotics, this problem can be easily handled. In this talk an introduction of some of the topics needed to tackle these problems will be presented.
Title: Model-based control of manipulators and walking robots with (high-)compliant series elastic actuators
Abstract: In this talk I will give insight into different control approaches that were applied to regulate ANYpulator and ANYmal – a 6DOF manipulator and a quadrupedal robot built form series elastic actuation modules. I will discuss to what level of detail compliant actuators need to be modeled to achieve good performance for model based control, MPC-style trajectory optimization, sampling-based trajectory optimization, and learning-based control.
Title: Soft robot bodies: a control challenge or a solution for robot control?
Abstract: Largely inspired by the observation of the role of soft tissues in living organisms, the use of soft matter for building robots is recognized as one of the current challenges for pushing the boundaries of robotics technologies and building robotic systems for service tasks in natural environments. But, in robotics, soft bodies pose important challenges for modelling and controlling their behavior, which are leading to the exploration of new approaches and to the development of novel techniques and tools. At the same time, the study of living organisms sheds light on principles that can be fruitfully adopted to control soft robot bodies. Living organisms exploit soft tissues and compliant structures to move effectively in complex natural environments. Their study can unveil the fundamental approaches that nature adopts to achieve such flexibility without the need of highly detailed internal models, but rather relying on the fundamental principles underpinning a specific behavior, with two general advantages. As the first one, selectively designed and embedded compliant components could be exploited to adopt self-stabilizing desired behaviors, eventually with advantages under the disturbance rejection, energy consumption and control-simplicity point of views. As second advantage, a soft body could be morphed to adopt different stances, shapes, or actuation patterns: where shape-dependent or environment-interaction forces are predominant, such capability could be desirable to resort to a morphological change the solution of a control problems. Despite the challenges in modelling and control of soft bodies, fundamental simplified models could actually capture such overall behavioral changes, actually leading the morphing of the body to obtain the desired behavior. Preliminary results in the control of soft robots, both with selective compliance or morphing body characteristics, are reported by following the approaches mentioned above. - Abstract is co-authored with Prof. Marcello Calisti, Scuola Superiore Sant'Anna in Pisa
Title: Screw Theory as a Unified Approach for Rigid, Soft Articulated and Soft Continuum Robots
Abstract: In the quest for advanced robots capable of agile and natural motion, researchers have developed new robotic designs which incorporate elastic, soft and highly deformable parts among their constituents. Those soft parts provide distributed degrees of freedom (DOFs) to the system and are physically characterized by distributed parameters, modelled by partial differential equations. On the contrary, conventional rigid robots display lumped DOFs and concentrated parameters, usually modeled with Euler-Lagrange methods. This discrepancy yielded to a deep hiatus between these two robotic disciplines, which is also spreading within the same soft robotics field depending on the percentage of rigid and elastic components in the system. Despite this blunt picture of the state-of-the-art, such a huge discrepancy between rigid and soft robots is not necessarily observed at the fundamental laws that govern the dynamics of the two kinds of systems. In this talk, a unified approach for modeling rigid, soft articulated and soft continuum robots is presented, which, making use of screw theory, includes lumped and distributed parameters/DOFs within the same framework without compromising on the richness provided by these novel advanced systems. Going further into the details, the Brockett’s geometric theory of conventional robotics is extended here to soft robotics by means of a discrete Cosserat modeling of the soft-body dynamics. We believe that the proposed unified theory may effectively bridge the gap between the different sub-disciplines that have raised with the development of elasticity-driven robots.
Title: Physics based modeling of deformable robots for real-time applications like simulation and control
Abstract: Continuum mechanics provide accurate mechanical models for deformable solids. Numerical tools, like the Finite Element Methods (FEM), have been developed to solve the partial differential equations. But a major drawback seems to prevent their use for real-time applications like robotics: FEM has the reputation to be slow. This presentation will show that there are some solutions to make FEM models fast enough to be compatible with real-time simulation and control methods. We will also show that this type of model can be mixed with more standard articulated and rigid models. Finally we will present the performance of this approach for modeling, simulation and control of soft-robots.
Title: Reduced order modeling of soft robots
Abstract: The passage from finite to infinite degrees of freedom in the modeling of robots (i.e., the passage from rigid to deformable parts) induces a considerable additional complexity. Usual analysis techniques in the field of continuum mechanics, such as the ubiquitous finite element method, are not thought to provide real-time response so as to be included in the control loop, unless stringent assumptions are made. In order to overcome these difficulties, model order reduction (MOR) techniques could be employed so as to simplify as much as possible the modeling of soft robots. These techniques allow to minimize the number of relevant degrees of freedom while keeping the accuracy of the model within a prescribed threshold. In this talk we have addressed the viability of employing MOR for the modeling and simulation of soft robots under stringent real-time constraints. The obtained results, still preliminary, show great promise, as will be demonstrated through the analysis of some examples.
Title: Linking an information transfer network and memory in soft body dynamics
Abstract: Behaviors of robots are generated by the dynamic coupling between their controller, body, and environment. Information-theoretic approach has been a useful tool to understand these coupling regimes in embodied robots. Directed information theory is particularly suitable for revealing causal relationships in terms of the strength between variables, as exemplified in transfer entropy or conditional mutual information. In this presentation, we show that the information transfer network topology revealed in soft body dynamics can be related to memory capacity in the sense of reservoir computing. Reservoir computing is a framework to exploit high-dimensional dynamical systems as a computational resource. Recently, we have shown that the diverse dynamics of a soft body can be exploited to embed nonlinear dynamical systems and a closed-loop control into its body. We demonstrate that two different fields, directed information theory and reservoir computing, are actually linked together within our approach and discuss its implication and potential in soft robot control.
Title: The Evolution and Development of Soft Robot Morphologies and Controllers
Abstract: While modeling soft robots to find the optimal control policies and body shapes for a given performance metric is a challenging engineering endeavor, biological optimization algorithms such as evolutionary adaptation, learning, and development have optimized both the morphology and control of soft and compliant plants and animals with amazing success. This talk will describe work using evolutionary algorithms (including the evolution of development) to design the shape and control of voxel-based simulated soft robots for a variety of tasks and environments. It will also highlight some of the challenges that arise when simultaneously co-evolving morphology and control, and describe the methodology for an algorithm which tackles some of these issues -- as well as examining and its theoretical underpinnings in embodied cognition. While the evolutionary population-based approach is just one of many possible methods for optimizing soft robot body plans and controllers, the black-box and derivative-free nature of this approach can provide some advantages that are especially practical in difficult to model systems such as these.