Elsevier

Neurocomputing

Volume 148, 19 January 2015, Pages 112-121
Neurocomputing

Co-evolution framework of swarm self-assembly robots

https://doi.org/10.1016/j.neucom.2012.10.047Get rights and content

Highlights

  • A co-evolution framework of configuration and control for swarm self-assembly robots is proposed.

  • We establish a unified control model to achieve diverse configurations of robot organisms in co-evolution process.

  • Simulation platform based on realistic robots is designed to implement the algorithm.

  • Co-evolution of configuration and control is proved to be successful on our simulation platform.

Abstract

In this paper, we present a co-evolution framework of configuration and control for swarm self-assembly robots, Sambots, in changing environments. The framework can generate different patterns composed of a set of Sambot robots to adapt to the uncertainties in complex environments. Sambot robots are able to autonomously aggregate and disaggregate into a multi-robot organism. To obtain the optimal pattern for the organism, the configuration and control of locomoting co-evolve by means of genetic programming. To finish self-adaptive tasks, we imply a unified locomotion control model based on Central Pattern Generators (CPGs). In addition, taking modular assembly modes into consideration, a mixed genotype is used, which encodes the configuration and control. Specialized genetic operators are designed to maintain the evolution in the simulation environment. By using an orderly method of evaluation, we can select some resulting patterns of better performance. Simulation experiments demonstrate that the proposed system is effective and robust in simultaneously constructing the adaptive structure and locomotion pattern. The algorithmic research and application analysis bring about deeper insight into swarm intelligence and evolutionary robotics.

Introduction

In nature, biological systems consisting of vast numbers of simple agents can attain functionally rich collective behavior and show impressive collective problem-solving capabilities [1], e.g., ant colonies, schools of fish, and multicellular organisms. The agents of nature allow cooperative and competitive working in large-scale societies. There exist two amazing collective phenomena [2]. First, the swarm agents can work collectively to profit from swarm capacity and intelligence, e.g., collective actuation, foraging, and exploration. Second, the swarm agents can aggregate into a multicellular organism that cannot be fulfilled by a single one or, in some cases, the swarms do not work collectively, e.g., in reaching target areas separated from the swarms by an object. Such biological phenomena have inspired the development of the robots, especially modular robots to improve adaptation.

We first present the swarm self-assembly robot, called Sambot, which has synthesized the strength of self-reconfigurable robots and self-assembly robots. Each Sambot is a completely autonomous mobile robot, similar to the individual robot in the swarm robotics. Multiple Sambots can form a robotic structure through self-assembly. A unified representation method is proposed to express the configuration. To derive the generic control model, we introduce the Central Pattern Generators (CPGs) by combining the previous unified representation method. Building upon the configuration and control model, a co-evolution framework is proposed to design the organism and the corresponding locomotion pattern that is composed of a swarm of robots. The algorithmic contributions of this work leads to a generic framework that co-evolve the configuration and control to allow the robot swarms forming the diverse patterns. In addition, we also present empirical results from implementing this algorithmic framework on a simulation experiment on co-evolution of the organism and locomotion control. Our proposed co-evolution framework is closely related to the evolutionary algorithm in modular robots [2], [3], [4]. Our work differs from other designs in the following two ways: (1) In the algorithmic aspect, we propose a generalized configuration and control model to give a unified genotype representation of all of the organisms even as the special operators are generating evolution. (2) In the aspect of application, the whole framework is established on existing swarm self-assembly systems. The systemic modules can autonomously aggregate and disaggregate to achieve a variety of organisms, which give the robot the potential to solve such issues of evolutionary organisms as configuration pattern, assembly, control, and encoding.

The rest of the paper is organized as follows: we begin with a related work review in Section 2. Section 3 introduces the Sambot simulation platform based on a realistic module, followed in Section 4 by a representation of the symbiotic organism and a unity locomotion control model; Section 5 describes the co-evolutionary framework for achieving an adaptive combination of configuration and control to carry out an intended task; Section 6 demonstrates the simulation of the organism׳s adaptation; and finally, Section 7 concludes the paper.

Section snippets

Related work

The co-evolution of morphology and controllers has proved to be successful for performing particular tasks [5], [6]. A genetic language using a directed graph of nodes and connections can define the nature of the creature. Virtual creatures are then expressed in a virtual environment to achieve fitness. These interesting virtual creatures interact with the external environment and then the best is selected. It is noted that in dynamic evolution a complex genotype-fitness relationship is

Sambot robot

Sambot is an autonomous mobile and self-assembly modular robot that can form a symbiotic or multicellular organism by modules moving autonomously and docking with each other [13], [14], [15]. The overall size of the module is 80×80×102 mm. As an autonomous independent unit like a cell, each module is homogenous and possesses the following capabilities:

  • (1)

    Sensor. Each robot module is equipped with a number of infrared sensors to measure distance, as shown in Fig. 1 ; gyroscope to measure direction,

Configuration topology

As part of the genetic representation, the topology of the configuration should be described as encompassing the characteristics of the module. In some evolutionary research on robots, different representations have been proposed, such as a directed graph [5], a tree-based type [10], an L-system [8], and an incidence matrix [3]. These representations can be divided into two types: direct representations and generative representations. The former is straightforward and coherent, while the latter

Co-evolution model

We focus on the dynamic coupling of the control system and the structural configuration with outside environments. Evolution [23] and swarm intelligence [1] can provide efficient method. An evolutionary algorithm that considers both robot configuration and control parameters is introduced.

Each assembly robot with a certain configuration has a unique chromosome. An initial population composed of robots with different chromosomes is generated at random. Chromosome encoding is composed of

Experiments and discussion

A co-evolutionary algorithm framework is designed to guide swarm self-assembly robots into an adaptive organism. Here, some experiments, including a dynamic simulation of the formation and locomotion of organisms, are demonstrated.

Conclusions

In this paper, an evolutionary framework was introduced to the Sambot platform, inspired by the evolution of biological organisms. The Sambot׳s representation and control model is generic enough to cover a large diversity of organism topologies. Taking genetic encoding into consideration, genetic operators using rule-based methods are presented. The co-evolution process is directed by the structure׳s locomotion capacity. Hence, it is necessary to ensure the structure׳s “fitness” to

Acknowledgments

This work was supported by the National High Technology Research and Development Program of China (863 Program) (2012AA041402), National Natural Science Foundation of China (Grant no. 61175079 and no. 51105012).

Haiyuan Li was born in Shandong Province, China, in 1986. He received a B.E. degree from Shandong University, Jinan, China in 2005, and an M.S. degree from Beihang University, Beijing, China, in 2012. He is currently working toward a Ph.D. in the School of Mechanical Engineering and Automation, Beihang University, Beijing, China. His current research interests include self-assembly swarm robots, modular robotics architecture, mobile sensor networks, and embedded systems.

References (24)

  • H. Lipson, J. Pollack, Towards continuously reconfigurable self-designing robotics, in: Proceedings of IEEE...
  • D. Marbach, A. Ijspeert, Co-evolution of configuration and control for homogenous modular robots, in: Proceedings of...
  • Haiyuan Li was born in Shandong Province, China, in 1986. He received a B.E. degree from Shandong University, Jinan, China in 2005, and an M.S. degree from Beihang University, Beijing, China, in 2012. He is currently working toward a Ph.D. in the School of Mechanical Engineering and Automation, Beihang University, Beijing, China. His current research interests include self-assembly swarm robots, modular robotics architecture, mobile sensor networks, and embedded systems.

    Hongxing Wei was born in the Inner Mongolia Autonomous Region, China, in 1974. He received a Ph.D. from the College of Automation, Harbin Engineering University, Harbin, China, in 2001. Since 2004, he has been an Associate Professor in the School of Mechanical Engineering and Automation, Beihang University (formerly Beijing University of Aeronautics and Astronautics), Beijing, China. His current research interests include self assembly swarm robots, modular robotics architecture, mobile sensor networks, and embedded systems.

    Jiangyang Xiao was born in Jiangsu Province, China. He received a B.E. degree and a M.S. degree from Ordnance Engineering College, Shijiazhuang, China, respectively in 1995 and 2001. Now, he is a senior engineer in Beihang University, Beijing, China. His current research interests include automation and robotics.

    Tianmiao Wang received a B.E. degree from Xi׳an Jiaotong University, Xi׳an, China, in 1982, an M.S. and Ph.D. from the Northwestern Polytechnical University, Xi׳an, in 1990 and 1997, respectively. He was a Postdoctoral Fellow in the State Key Laboratory of Intelligent Technology and Systems, Tsinghua University, in 1992, and at the State Bionic Force Laboratory, Italy, in 1995. Since 1998, he has been a Professor in the School of Mechanical Engineering and Automation, Beihang University (formerly Beijing University of Aeronautics and Astronautics), Beijing, China. His research interests include mircorobot technology, medical robot technology, and embedded electromechanical control technology.

    View full text