Multi-robot localization and orientation estimation using robotic cluster matching algorithm

https://doi.org/10.1016/j.robot.2014.09.002Get rights and content

Highlights

  • A new algorithm for multi-robot localization and orientation is introduced.

  • Clusters of nodes scanned by distance IR sensor to estimate their location and orientation.

  • The location and orientation of invisible robots are computed by using location estimation algorithm.

  • Several simulation scenarios are implemented to indicate the performance of suggested algorithm.

Abstract

In this paper, a new algorithm, called cluster matching, is introduced for multi-robot localization and orientation. This algorithm deals with the case in which each robot has the capability to estimate the relative orientation of those robots (called neighbors) that are within its transmission range. Furthermore, the environment is equipped with a distance IR sensor scanning the robots and estimating the absolute positions and orientations of a number of the team robots without knowing their IDs. The IDs of these robots are reconstructed by matching the orientation obtained by the distance IR sensor with the relative orientation measured with on-board sensors. The localization and orientation of robots not visible to the distance IR sensor are obtained by collecting the information coming from the on-board sensors and thus reconstructing a complete map of the team distribution. The accuracy in the estimation of the location of these robots is enhanced by introducing a new algorithm which relies on the localization of neighbor robots. Several simulation scenarios are implemented on tens of robots to show the performance of the introduced algorithm.

Introduction

Multi-robot systems have several advantages over single robot systems. They can localize themselves more efficiently  [1], improve the speed of search and exploration missions  [2], [3], accomplish task in shorter times, and also increase accuracy and fault tolerance. Multi-robot systems consist of simple homogeneous robots that have low computational capability due to cost constraints  [4], [5]. The problem of localization, that is, the determination of the robot position in a given map of the environment, is a central issue of mobile robotics in general and, in particular, of multi-robot systems. Localization may be absolute or relative  [6]. Localization is absolute when makes use of landmarks, global maps, beacons, or satellite signals to determine the position and orientation of the robot. Absolute localization methods require that the environment is known or mapped with great accuracy. Methods based on satellites, such as for instance GPS, can be used only with outdoor environment and have poor accuracy in indoor environment. Relative localization is usually preferred during movement, because absolute localization methods are more time consuming. One widely used method for relative positioning is for instance odometry (that is, the distance from a starting location is computed by measuring the wheel revolutions). This is a simple and easy to achieve method, which, however, suffers from cumulative errors.

The problem of localization is clearly connected to the choice of the map used to represent the environment. Maps are classified into two categories, referred to as feature-based maps and location-based maps. Feature-based maps represent the environments as a set of features with their Cartesian location, such as the lines and corners in the environment  [7], [8], [9]. Location-based maps produce a marker for any location in the environment; a classical type of this map is the occupancy grid  [10], [11]. In many cases, a priori maps of the environment are not available. In such cases, when the robotic system is made by a team of robots, localization and mapping can be simultaneously and efficiently addressed. In fact, each robot of the team can provide partial information on the environment being explored as well as information on the relative positions of the other robots which are at a small distance from it and are usually referred as neighbors   [12], [13], [14]. These methods, therefore, exploit the sensing/communication capabilities of the robots in the team which makes them able to sense the other robots and share the data with them  [15], [16] to generate a number of local maps of the environment. Such maps then need to be integrated to generate the global map.

Topological partial maps, created by using the limited range sensors equipped on individual robots, may for instance be merged by using global reference frame to construct the global map  [17], while, for instance, landmark-based maps may be merged into a single map, based only on the feature sets that each robot independently discovers  [18]. Local maps integration may be performed also by using a localization method based on multidimensional scaling (MDS), such as MDS-MAP  [19], where the maps represent the distribution of the robots, also viewed as nodes of a network (being the links representative of a neighborhood relationship between two robots). Each robot generates a local map, which thus represents the distribution in the environment of the other robots. Then, when the positions of a sufficient number of anchor nodes are known, MDS-MAP becomes able to determine the absolute coordinates of all robots in the network.

The problem of integration of local maps is related to that of matching up successive sets of raw range readings, called scans, an issue arising when the robot displacement has to be determined. For this reason, methods based on scan matching   [20], [21], [22] may be also used to generate a global map from a set of local ones.

Scan matching may be performed with different types of sensors: either laser sensors providing thousands of readings per second with a subdegree angular resolution or low cost distance IR sensors, more suited for being equipped on mobile robots due to their size, price and power consumption, are used. The algorithms based on scan matching usually make use of the concepts introduced in the Iterative Closest Point (ICP) approach  [23]. The ICP algorithm makes possible to build local maps for sequential positions of robot while exploring the environment, and successively, to calculate the robot positions by matching the local maps  [24]. New studies verified that ICP-based algorithms can be used with accurate sonar sensors  [20], [25]. In an ICP-based framework, each sonar reading is modeled as a normal distribution. Geometric compatibility tests are used to find reading to reading correspondences [26]. In order to apply the ICP algorithm efficiently, a searching algorithm is used to identify the potential common part of two successive frames of sensor data  [27].

This paper aims to introduce a localization and orientation system combining a source of absolute, but partial information, that is, a distance IR sensor anchored in a fixed frame of the environment, and a source of relative (and partial) information, that is, the local maps generated by the robots of the team while exploring the environment. The approach is designed for a multi-robot system, made of n units with directional infrared communication links. More in detail, each robot is equipped with m directional infrared communication channels distributed uniformly around the circumference of the robot. Each infrared sensor is capable to communicate with the corresponding sensors of the neighboring robots. The distance IR sensor, placed in a corner of the environment, is used to scan the environment and measure the distances between itself and robots. From these sources two maps of the robot distribution in the environment are obtained; as in the MDS-MAP approach, both maps are here represented as networks where the nodes are the robots and a link between two nodes exists if the robots are neighbors. From the distance IR sensor we obtain the cluster network, while from the integration of the local maps from the on-board sensors we obtain a unit disk graph. In the cluster network only a subset of the robots will be represented as some robots may be not visible for the distance IR sensor, but for all the robots in this network an absolute distance will be available, as the measure of the distance IR sensor is an absolute one. In the unit disk graph, on the contrary, only relative distances will be known. Even in the case of the unit disk graph not all the robots of the team will be represented as some of them may be at a distance greater than the range of the on-board sensors of all the other robots of the system. The two maps are then merged so that absolute localization may be performed for a number of robots higher than those belonging to the cluster network. An overall view of the whole system is shown in Fig. 1, with the indication of the algorithms used to process the data coming from the two sources of information. These algorithms are discussed in detail in Section  2. Section  3 describes the simulation results of the robotic clusters matching algorithm. Finally, Section  4 draws the conclusions of the paper.

Section snippets

The robotic cluster matching algorithm

In this section, the new algorithm for multi-robot localization referred as robotic cluster matching algorithm is introduced. This algorithm determines the position and orientation of the robots with respect to a global reference frame. The localization problem is addressed in two stages: in the first stage, it is assumed that the absolute positions of some robots are known (from the distance IR sensor) but the information is limited to those robots visible to the distance IR sensor. These

Simulation results

The robotic cluster matching algorithm is validated using simulations. It is implemented using visual basic 2010. Initially, a unique node ID is assigned to each robot. Simulations are performed over 150 different topologies representing different network sizes n ranging from 5 to 100 robots. The robots are randomly placed according to a uniform distribution on a 500×500 pixels area. For each topology, the transmission range of each robot R is varied in order to study the effect of different

Conclusions

In this paper, a new algorithm for multi-robot localization and orientation has been introduced. This algorithm is based on the idea of matching information from different sources (a fixed frame and on-board sensors) in order to perform localization of a multi-robot system. The fact that the information to be matched comes from different sources is a key feature of our algorithm as the matching is usually done on different readings obtained from the same sensor or from the same type of sensors

Abdulmuttalib T. Rashid was born in Iraq. He received the B.S. degree in Electrical Engineering from Basrah University at Basrah, Iraq in 1986. He received the M.Sc. degree from the same university in 1992. He has worked as an Assistant Lecturer and a Lecturer in the Department of Electrical Engineering, University of Omer Al Mukhtar, Lybia from 1997 to 2007. He works in the Department of Electrical Engineering, University of Basrah, Iraq since 2007. His field of interest is Robotics and

References (31)

  • A. Marjovi et al.

    Multi-robot olfactory search in structured environments

    Robot. Auton. Syst.

    (2011)
  • K. Leung et al.

    Decentralized localization of sparsely communicating robot networks: a centralized-equivalent approach

    IEEE Trans. Robot. Autom.

    (2010)
  • A. Marjovi, J. Nunes, P. Sousa, R. Faria, L. Marques, An olfactory-based robot swarm navigation method, in: Proc. IEEE...
  • A. Marjovi, L. Marques, Multi-robot topological exploration using olfactory cues, in: Int. Symp. on Distributed...
  • C. Yu et al.

    A self-adaptive framework for modular robots in dynamic environment: theory and applications

    Int. J. Robot. Res.

    (2010)
  • J. Borenstein et al.

    Measurement and correction of systematic odometry errors in mobile robots

    IEEE Trans. Robot. Autom.

    (1996)
  • J.A. Castellanos et al.

    Multi sensor fusion for simultaneous localization and map building

    IEEE Trans. Robot. Autom.

    (2001)
  • G. Dissanayake et al.

    Map management for efficient simultaneous localization and mapping (SLAM)

    Auton. Robots

    (2002)
  • M. Bosse et al.

    SLAM in large-scale cyclic environments using the Atlas framework

    Int. J. Robot. Res.

    (2004)
  • H.P. Moravec

    Sensor fusion in certainty grids for mobile robots

    AI Mag.

    (1988)
  • A. Elfes

    Occupancy grids: a probabilistic framework for mobile robot perception and navigation

    (1989)
  • P. Biswas, Y. Ye, Semidefinite programming for ad hoc wireless sensor network localization, in: Proceedings of the 3rd...
  • D. Niculescu, B. Nath, Error characteristics of ad hoc positioning systems (APS), in: MobiHoc’04: Proceedings of the...
  • Y. Shang, W. Ruml, Y. Zhang, M.P.J. Fromherz, Localization from mere connectivity, in: MobiHoc’03: Proceedings of the...
  • S.I. Roumeliotis et al.

    Distributed multi-robot localization

    IEEE Trans. Robot. Autom.

    (2002)
  • Cited by (0)

    Abdulmuttalib T. Rashid was born in Iraq. He received the B.S. degree in Electrical Engineering from Basrah University at Basrah, Iraq in 1986. He received the M.Sc. degree from the same university in 1992. He has worked as an Assistant Lecturer and a Lecturer in the Department of Electrical Engineering, University of Omer Al Mukhtar, Lybia from 1997 to 2007. He works in the Department of Electrical Engineering, University of Basrah, Iraq since 2007. His field of interest is Robotics and Industrial control. Currently, he is pursuing his Ph.D. in Electrical Engineering in the University of Basrah, since 2009. His research interests include motion planning and control of multi mobile robots.

    Mattia Frasca was born in Siracusa, Italy, in 1976. He graduated in Electronics Engineering in 2000 and received the Ph.D. in Electronics and Automation Engineering in 2003, from the University of Catania, Italy. Currently, he is a Research Associate at the University of Catania. His scientific interests include nonlinear systems and chaos, Cellular Neural Networks, complex systems and bio-inspired robotics. He is involved in many research projects and collaborations with industries and academic centers. He is referee for many international journals and conferences. He was in the organizing committee of the 10th “Experimental Chaos Conference” and co-chair of the “4th International Conference on Physics and Control”. He is coauthor of three research monographs (with World Scientific): one on locomotion control of bio-inspired robots, one on self-organizing systems and one on the Chua’s Circuit. He has published more than 150 papers in refereed international journals and international conference proceedings and is co-author of two international patents. He is an IEEE senior member.

    Abduladhem A. Ali received his M.Sc. and Ph.D. degrees from the Department of Electrical Engineering, University of Basrah, Iraq in 1983 and 1996. He has worked as an Assistant Lecturer, a Lecturer, and as an Assistant professor in the same department in 1984, 1987 and 1981 respectively. Then he has worked as an Assistant Professor and a Professor in the Department of Computer Engineering in 1997 and 2004 respectively. He has worked as a consultant to many industrial firms to design industrial control systems. He has more than 70 published papers, one patent, and has supervised many M.Sc. and Ph.D. dissertations. He is the Editor chair for the Iraqi Journal for Electrical and Electronic Engineering and a member of the editorial board for many Journals. He was the Chairman of the first IEEE International conference on Energy, Power and Control (EPC-IQ01). His field of interest is Robotics, Industrial control and Intelligent systems. He is currently the Director of Avicenna E-learning Center at the University of Basrah, Iraq.

    Alessandro Rizzo is a Tenured Assistant Professor in Automation Engineering. His research interests are focused on complex networks and systems, distributed estimation and control, bio-inspired control and robotic systems, soft sensors. From September 1, 2013 to May 31, 2014 he was at the Polytechnic Institute of New York University, as a Visiting Professor at the Dynamical Systems Lab of the Department of Mechanical and Aerospace Engineering.

    Luigi Fortuna (M’90–SM’99–F’00) was born in Siracuse, Italy, in 1953. He received the degree of Electrical Engineering (cum laude) from the University of Catania, Catania, Italy, in 1977. He is a Full Professor of system theory with the Università degli Studi di Catania, Catania, Italy. He was the coordinator of the courses in Electronic Engineering and the Head of the Dipartimento di Ingegneria Elettrica Elettronica e dei Sistemi (DIEES). Since 2005, he has been the Dean of the Engineering Faculty, Catania. He currently teaches complex adaptive systems and robust control. He has published more than 450 technical papers and is the coauthor of ten scientific books, among which are: Chua’s Circuit Implementations (World Scientific, 2009), Bio-Inspired Emergent Control of Locomotion Systems (World Scientific, 2004), Soft-Computing (Springer 2001), Nonlinear Non Integer Order Circuits and Systems (World Scientific 2001), Cellular Neural Networks (Springer 1999), Neural Networks in Multidimensional Domains (Springer 1998), Model Order Reduction in Electrical Engineering (Springer 1994), and Robust Control—An Introduction (Springer 1993). His scientific interests include robust control, nonlinear science and complexity, chaos, cellular neural networks, soft computing strategies for control, Robotics, micronanosensor and smart devices for control, and nanocellular neural networks modeling.

    He was the IEEE Circuits and Systems (CAS) Chairman of the CNN Technical Committee, IEEE CAS Distinguished Lecturer from 2001 to 2002, and IEEE Chairman of the IEEE CAS Chapter Central-South Italy.

    View full text