Roombots: A hardware perspective on 3D self-reconfiguration and locomotion with a homogeneous modular robot
Introduction
We are working towards the idea of a living environment where classic roomware components are merged and enhanced by elements from robotics and information technology. We tackle this task by designing Roombots (RB), an instance of self-reconfiguring modular robots (SR-MR) which are meant to be embedded into our living environment, to build intelligent furniture and other components of our daily life. Self-reconfiguring modular robots are highly integrated, self-sustaining robotic building blocks with limited degrees of freedom (DOF). A number of them can connect to each other with active connection mechanisms (ACM), creating a robot or structure with more capabilities than a single module. Eventually, we see the task of RB to build adaptive furniture that connects into different shapes, such as stools, tables, or sofas. It should also adapt to the user’s needs, by changing shape over time, and by moving around.
Several designs of self-reconfiguring modular robotic systems have been proposed, with the goal to assemble larger structures [1], [2]. Such structures could be complex assemblies, such as parts of a space station, remote, hard-to-reach and hard-to-maintain systems such as a modular satellite, deep-sea underwater structures, or scaffolds. For the Roombots project, we are aiming to assemble furniture-like structures. For our furniture, we want to minimize the number of active robotic modules by combining active modules and passive elements—a table assembled from light-weight elements and a few Roombots modules would be by far cheaper, less heavy and structurally more sound than one made entirely from modular robots. Currently, we envision a mix of Roombots modules and passive elements, as depicted in Fig. 1. Light-weight elements of several shapes and Roombots modules would create a patchwork, held together by Roombots modules. Connector “ports” play an important role in this scenario; these are distributed around the room as passive connectors in the floor, walls, ceilings, and within the light-weight elements. The same ports are also part of Roombots modules, as its active and passive connectors.
Towards this future goal, we tackle and present experimental results for five sub-scenarios in this work. Firstly, we imagine a stock of Roombots modules on the side, and modules detaching from it. Not all the floor will be equipped with connector ports. To reach a port-equipped area (“on-grid”), we show how pair-wise connected Roombots metamodules (MM) locomote with oscillating or wheel-like rotating motion patterns over the floor (“off-grid” locomotion, Fig. 1, subtask 1). Module locomotion is controlled by a central pattern generator (CPG) [3] implemented as a network of coupled oscillators, distributed in the different modules. CPG are decentralized and well suited to organize rhythmic and non-rhythmic motions of large numbers of modular robots.
Next, modules have to become aligned to the grid. We propose a passive mechanism based on an entrapment-like structure (Fig. 1, subtask 2): Roombots metamodules slide into a sink-like entrapment mechanism (EM). In there the MM automatically aligns, the MM’s ACM grippers connect to a EM wall port. With a couple of predefined joint moves the MM leaves the EM, and “brachiates” over the grid connection ports [4], [5] (simulation results). Roombots modules can perform on-grid locomotion by iteratively attaching and detaching to the ports. In some cases, modules fail to attach when controlled in open-loop because of excessive bending of the structure, and necessary hardware updates for this task are discussed in Section 5.
To build a larger structure from modules and passive elements, Roombots modules will have to approach, climb, and overcome walls and planes. Those obstacles present themselves either as “concave” edges, planes, or “convex” edges. Single Roombots modules are designed and controlled to overcome concave edges (Fig. 1, subtask 3), and to switch from a horizontal plane to a vertical plane. To go upwards and switch to the next-level horizontal plane a single Roombots module has to be merged into a full Roombots metamodule, together with a Roombots “helper module” (Fig. 1, subtask 4).
Roombots are meant to assemble larger structures from passive elements. In subtask 5 we present a brief first step towards handling light-weight structures. In fact, the future Roombots hardware will require a stiffness and power upgrade before being able to handle larger elements, as depicted on the left side of the figure. For now, two Roombots modules are picking up a pair of light-weight connector plates (Fig. 1, similar to 5). For the future of this project, remaining challenges will include the tasks of picking up light-weight elements by cooperating Roombots metamodules, transporting those elements to their assembly point (Fig. 1, two metamodules lifting a light-weight element to the table top), and finally mounting and assembling everything into meaningful structures. In this work we focus at the defined five subtasks and the necessary modular robot hardware to robustly solve these scenarios. Remaining challenges (e.g. multi-metamodule handling of passive elements, RB metamodule locomotion on-grid) are identified along with this work, and are analyzed and discussed in Section 5.
The paper is structured as follows: Section 2 gives a brief overview over modular robot (MR) classifications, and significant MR implementations. In Section 3 we present the mechanical, electrical, and control design of Roombots. In Section 4, we show five hardware experiments of reconfiguration and locomotion on-grid and off-grid, in 2D and in 3D, alignment of RB from off-grid to on-grid, and handling of light-weight elements through RB modules. We discuss our results and future work in Section 5, and conclude the paper in Section 6.
Section snippets
Related work
In this section, we survey and analyze a selection of modular robots, with a focus on module density, weight, and classification of modular robots. The background of this survey is the Roombots scenario, and the core tasks of RB modules (a) to attach and detach robustly with other RB modules and the environment, (b) to locomote with brachiating movements through the surrounding on-grid environment and move in the off-grid environment and (c) to handle other RB modules and light-weight elements.
Material and methods
This section gives details of Roombots’ mechanical design (Section 3.1). This includes its active connection mechanism (ACM), actuator design, and shell and grid design. Section 3.2 presents Roombots’ electrical components. Section 3.3 describes control schemes for distributed locomotion, and reconfiguration.
Experimental results
To demonstrate the capabilities of the RB system, we conducted five main experiments. The results are presented in this section. Video including all experiments is given below:
Discussion and future work
The presented hardware experiments show the capabilities of the current Roombots generation. These experiments also reveal the remaining challenges for the full, envisioned Roombots scenario. We successfully demonstrated that RB modules autonomously move on a 2D grid, overcome concave corners, form metamodules, and cooperate to overcome convex corners in a structured 3D-grid environment. We presented a proof-of-concept showing how RB modules can handle passive, light-weight parts. An entrapment
Conclusion
In this work we presented the design of the Roombots (RB) self-reconfiguring modular robotic system. In hardware experiments, we demonstrated locomotion and reconfiguration of single RB modules and RB metamodules (two modules connected in-series) in a 3D structured grid environment. Modules performed on-grid locomotion by “brachiating” along connector ports, overcame concave and convex edges, and climbed up structured walls. RB assemblies also locomoted off-grid, in the non-structured
Acknowledgments
We gratefully acknowledge the help of Soha Pouya, Ebru Aidin, Manon Picard, Alexandre Tuleu, Alessandro Crespi, Christophe Chariot, Jocelyn Lotfi, Simon Lepine, Emilie Badri, Philippe Laprade, Thanh-Khai Dinh, Anh The Nguyen, Frédéric Wilhelm, Manuel Stöckli, Yurij Perov, Efthymios Stavridis, Peter Loeppelmann, Jesse van den Kieboom, and Mikaël Mayer for the Roombots project. We gratefully acknowledge the technical support of André Guignard, André Badertscher, Philippe Vosseler, Mitchell
Alexander Spröwitz is a post-doctoral researcher. He worked until 2013 at the Biorobotics Laboratory, EPFL, Switzerland. He has a “Vordiplom” (B.Sc.) in Mechanical Engineering, and a “Diplom” (M.Sc.) in Biomechatronics from the University of Ilmenau, Germany, and a Ph.D. in Manufacturing Systems and Robotics from EPFL, Switzerland. His research includes design and control of self-reconfiguring modular robots and bipedal and quadrupedal legged robots, and biomechanics of legged robotic and
References (121)
Central pattern generators for locomotion control in animals and robots: a review
Neural Networks
(2008)Using cellular automata and gradients to control self-reconfiguration
Robotics and Autonomous Systems
(2006)- et al.
Concept of cellular robotic system (CEBOT) and basic strategies for its realization
Computers & Electrical Engineering
(1992) - et al.
A rapidly deployable manipulator system
Robotics and Autonomous Systems
(1997) - et al.
Living machines
Robotics and Autonomous Systems
(1995) - Y. Terada, S. Murata, Automatic assembly system for a large-scale modular structure: hardware design of module and...
- et al.
Planning the reconfiguration of grounded truss structures with truss climbing robots that carry truss elements
- A. Spröwitz, Roombots: design and implementation of a modular robot for reconfiguration and locomotion, Ph.D. Thesis,...
- S. Bonardi, R. Moeckel, A. Spröwitz, M. Vespignani, A. Ijspeert, Locomotion through reconfiguration based on motor...
- T. Fukuda, Y. Kawauchi, Cellular robotic system (CE-BOT) as one of the realization of self-organizing intelligent...
Modular self-reconfigurable robot systems: challenges and opportunities for the future
IEEE Robotics and Automation Magazine
Robotics: self-replication from random parts
Nature
Folding DNA to create nanoscale shapes and patterns
Nature
Multimode locomotion via SuperBot reconfigurable robots
Autonomous Robots
Learning to move in modular robots using central pattern generators and online optimization
International Journal of Robotics Research
Distributed self-reconfiguration of M-TRAN III modular robotic system
The International Journal of Robotics Research
Swarm-Bot: a new distributed robotic concept
Autonomous Robots
Approach to the dynamically reconfigurable robot systems
Journal of Intelligent Robtics Systems
Water floating self-assembling agents
Resilient machines through continuous self-modeling
Science
Crystalline robots: self-reconfiguration with compressible unit modules
Autonomous Robots
Physical connection and disconnection control based on hot melt adhesives
IEEE/ASME Transactions on Mechatronics
Shady: robust truss climbing with mechanical compliances
Self-reconfiguring robots
Communications of the ACM Archive
Self-reproducing machines
Nature
M-TRAN: self-reconfigurable modular robotic system
IEEE/ASME Transactions on Mechatronics
Multimode locomotion via SuperBot reconfigurable robots
Autonomous Robots
Roombots: reconfigurable robots for adaptive furniture
IEEE Computational Intelligence Magazine
Miche: modular shape formation by self-disassembly
The International Journal of Robotics Research
Cited by (96)
The internet of modular robotic things: Issues, limitations, challenges, & solutions
2023, Internet of Things (Netherlands)Autonomous distributed system for single-legged modular robots to traverse environments by adaptive reconfiguration
2022, Robotics and Autonomous SystemsCitation Excerpt :Surprisingly, ants do not have any leader to manage the global morphology, but achieve these tasks by assembling into an appropriate morphology through local information exchange. Achieving a task by self-assembly is also beneficial for artificial reconfigurable modular robot systems, which are composed of repeated robot elements with a mutual physical connection mechanism [6–10]. It is difficult for a robot with a constant morphology to achieve several tasks, but reconfigurable modular robots can reconfigure the morphology in response to the task.
Engineering efficient and massively parallel 3D self-reconfiguration using sandboxing, scaffolding and coating
2021, Robotics and Autonomous SystemsCitation Excerpt :Accordingly, discarding modules from the initial configuration would mean that these modules cannot be summoned back into the reconfiguring robot later without external communication means. Some authors have thus relied on wireless communication for that purpose, or structured (grid-like) environments supplying energy and used as a global communication bus between modules across the grid [33]. In a nutshell, hyponumeric reconfiguration thus requires the extra modules can be discarded somewhere near the reconfiguration scene, and hypernumeric reconfiguration requires a reserve of modules from which additional modules can be attracted.
Snail robots as a new paradigm for swarm robotics in unstructured environments
2023, Research Square
Alexander Spröwitz is a post-doctoral researcher. He worked until 2013 at the Biorobotics Laboratory, EPFL, Switzerland. He has a “Vordiplom” (B.Sc.) in Mechanical Engineering, and a “Diplom” (M.Sc.) in Biomechatronics from the University of Ilmenau, Germany, and a Ph.D. in Manufacturing Systems and Robotics from EPFL, Switzerland. His research includes design and control of self-reconfiguring modular robots and bipedal and quadrupedal legged robots, and biomechanics of legged robotic and animal locomotion, such as in running birds.
Rico Moeckel is a post-doctoral researcher. He received a Ph.D. degree from the Swiss Federal Institute of Technology in Zurich, Switzerland, and a diploma (equivalent to M.Sc.) in Electrical Engineering from the University of Rostock, Germany. His research interests are in the fields of legged and modular robotics, computational neuroscience, locomotion control and VLSI vision sensors.
Massimo Vespignani is a Ph.D. student in the Biorobotics Laboratory (BioRob) at EPFL. He received his M.S. degree in Biomedical Engineering from the Campus Bio-Medico University (Rome, Italy) in 2009. His research interests include reconfiguring modular robotic systems, mechanical design, bio-inspired robotics, legged locomotion and soft robotics.
Stephane Bonardi received a M.Sc. (engineering diploma) in Applied Mathematics and Modeling from the Institut des Sciences et Techniques de l’Ingenieur de Lyon (ISTIL-EPU) in 2009. He joined the Biorobotics Laboratory as a Ph.D. student in February 2010. He is working on the Roombots project, mainly on the aspects of self-reconfiguration and end-user interface.
Auke Jan Ijspeert is an associate professor at the EPFL (Ecole Polytechnique Federale de Lausanne, Switzerland), and head of the Biorobotics Laboratory. He has a “diplome d’ingenieur” in physics from the EPFL, and a Ph.D. in artificial intelligence from the University of Edinburgh. His research interests are at the intersection between robotics, computational neuroscience, nonlinear dynamical systems, and applied machine learning. He is interested in using numerical simulations and robots to get a better understanding of the sensorimotor coordination in animals, and in using inspiration from biology to design novel types of robots and adaptive controllers (see for instance Ijspeert et al. Science, vol. 315: 5817, pp. 1416–1420, 2007). For more information see: http://biorob.epfl.ch.