Multiple mobile robots system with network-based ...

Int. J. Mechatronics and Manufacturing Systems, Vol. 6, No. 1, 2013

Multiple mobile robots system with network-based subsumption architecture Fusaomi Nagata* and Akimasa Otsuka Department of Mechanical Engineering, Faculty of Engineering, Tokyo University of Science, 1-1-1 Daigaku-Dori, Sanyo-Onoda 756-0884, Yamaguchi, Japan E-mail: [email protected] E-mail: [email protected] *Corresponding author

Keigo Watanabe Department of Systems Engineering, Faculty of Engineering, Okayama University, 3-1-1 Tsushima-naka, Kita-ku, Okayama 700-8530, Japan E-mail: [email protected]

Maki K. Habib Mechanical Engineering Department, School of Sciences and Engineering, American University in Cairo, 113, Kasr El Eini St., P.O. Box 2511, Cairo 11511, Egypt E-mail: [email protected] Abstract: In this paper, a unique multiple mobile robots system is proposed to enable engineering students and engineers in the field to efficiently learn subsumption architecture and develop swarm intelligence. The subsumption architecture is known as one of the behaviour-based artificial intelligence. Each of multiple mobile robots within the system has three wheels driven by DC motors and six position sensitive detector (PSD) sensors. Network-based subsumption architecture is considered to realise a schooling behaviour by using only information from the PSD sensors. Further, a server supervisory control is introduced for poor hardware platforms with limitations of software development, i.e., the mobile robots can only behave based on the most simply subdivided reaction behaviours, i.e., reflex actions, generated from agents. Experimental results show interesting behaviour among the multiple mobile robots, such as following, avoidance and schooling.

Copyright © 2013 Inderscience Enterprises Ltd.

57

58

F. Nagata et al. Keywords: multiple mobile robots; network-based subsumption architecture; mechatronics educational system; PSD sensor, behaviour-based artificial intelligence. Reference to this paper should be made as follows: Nagata, F., Otsuka, A., Watanabe, K. and Habib, M.K. (2013) ‘Multiple mobile robots system with network-based subsumption architecture’, Int. J. Mechatronics and Manufacturing Systems, Vol. 6, No. 1, pp.57–71. Biographical notes: Fusaomi Nagata is a Professor at Department of Mechanical Engineering, Faculty of Engineering, Tokyo University of Science, Yamaguchi, Japan. Akimasa Otsuka is a Research Assistant at Department of Mechanical Engineering, Faculty of Engineering, Tokyo University of Science, Yamaguchi, Japan. Keigo Watanabe is a Professor at Department of Systems Engineering, Faculty of Engineering, Okayama University, Japan. His research interests include stochastic adaptive estimation and control, robust control, neural network control, fuzzy control, genetic algorithms and their applications to the robotic control. Maki K. Habib is a Professor at the Mechanical Engineering Department, School of Sciences and Engineering, American University in Cairo, Egypt. His main area of research are focusing on human adaptive and friendly mechatronics, autonomous navigation, service robots and humanitarian demining, telecooperation, distributed teleoperation and collaborative control, wireless sensor networks and ambient intelligence, biomimetic robots.

1

Introduction

Up to now, many multiple mobile robots systems have been developed according to various objectives. First of all, this paper surveys some of the main papers in the field that are associated with actual experimental results presented in chronological order according to the year of publication. Parsons and Canny (1990) proposed an algorithm for planning the motions of several mobile robots which share the same workspace containing polygonal obstacles. Each robot has an ability of independent translational motion in two dimensions. The algorithm computes a path for each robot which avoids all obstacles in the workspace as well as the other robots. Kube and Zhang (1992, 1993) examined the problem of controlling multiple behaviour-based autonomous robots. Based on observations made from the study of social insects, they proposed five simple mechanisms used to invoke group behaviour in simple sensor-based mobile robots. They also constructed a system of five homogeneous sensor-based mobile robots with capability of achieving simple collective task. The proposed mechanism allows populations of behaviour-based robots to perform tasks without having centralised control or use of explicit communication. Noreils (1992) described architecture for cooperative and autonomous mobile robots. The cooperation is composed of two phases. The first phase is the collaboration where a task is decomposed into subtasks. The second phase is the coordination where robots actually coordinate their activities to fulfil

Multiple mobile robots system with network-based subsumption architecture

59

the initial task using the notion of coordinated protocols. This architecture showed benefits of modularity, robustness and programmability. Barman et al. (1993) developed an extensible facility for multiple mobile robots. The system consists of nine radio-controlled mobile robots, two CCD colour video cameras, a video transmitter and tuner, radio controllers, an image processing hardware and so on. A software for tracking control is described too. In addition, Azarm and Schmidt (1997) presented a novel approach to do conflict-resolution for multiple mobile robots. A framework for negotiation is developed by using the online motion planning, which permits quick decentralised and parallel decision-making. The key objective of the negotiation procedure is dynamic assignment of robot’s motion priorities. The basic performance is evaluated from experiments using only two mobile robots. Bennewitz and Burgard (2000) considered the problem of path planning for teams of mobile robots. It presented a decoupled and prioritised approach to coordinate the movements of the mobile robots in their environment. The proposed algorithm computes the paths for the individual robots in the configuration-time space. To estimate the risk of colliding with other robots, it uses a probabilistic model of the robots motions. Guo and Parker (2002) proposed a distributed and optimal motion planning algorithm for multiple robots, in which the computationally expensive problem was decomposed into two modules, i.e., path planning and velocity planning. The D* search method was applied in both modules, based on either geometric formulation or schedule formulation. The algorithm was implemented and demonstrated successfully in a group of Nomad 200 indoor robot. Parker (2003) outlined the project that demonstrated a team of more than 100 heterogeneous robots solving an indoor reconnaissance and surveillance task. The specific problem to be solved was shown with the robot team. Pimentel and Campos (2003) addressed the problem of multi-mobile robot cooperation with strict communication constraints which are considered indispensable for successful task execution. The problem is instantiated as a cooperative search and rescue task, and is modelled as a minimisation of an energy function which accounts for network connectivity, other relevant robot and task requirements in order to select locally optimal actions for each robot. Antonelli et al. (2007) presented two experimental case studies performed using multi-robot system made of six Khepera II mobile robots. The experiments were aimed at testing the performance and the robustness of a behaviour-based technique, called the null-space-based behavioural control (NSB). The NSB approach was developed to control a generic team of autonomous vehicles and it was implemented on a centralised architecture to control a platoon of autonomous mobile robots at a kinematic level. Also, the experimental validation of the NSB was presented focusing on the experimental details, in which the validation in presence of static and dynamic obstacles was evaluated with a team of grounded mobile robots (Antonelli et al., 2009). Antonelli et al. (2010) further analysed the flocking behaviour in a variety of conditions: with or without moving rendezvous point, in a two- or three-dimensional space and in presence of obstacles. The effectiveness of the proposed algorithm was shown through simulations and experiments using a team of differential-drive mobile robots. The subsumption architecture proposed by Brooks (1986) is one of typical behaviour-based artificial intelligence. The subsumption architecture has been successfully implemented in the controllers of various types of robots, where layers of control system are built to let a robot to operate at increasing levels of competence. Recently, many studies on mechatronics educational systems are being conducted.

60

F. Nagata et al.

However, there are few systems that can support engineers or students to learn the basic concept of behaviour-based artificial intelligence such as the subsumption architecture. Compared with conventional control approaches, the subsumption architecture has advantages with respect to simplicity of architecture, extendability of function and especially time cost for software implementation. Although an experimental lecture of mechanical engineering must be finished within given hours, the subsumption architecture enables students to conduct programming and debugging within a limited time given for software implementation. When multiple mobile robots system is designed to experimentally simulate swarm intelligence or to be used for educational purpose, there exists two serious requirements: The first is to suppress the total cost associated with the rise of the number of mobile robots and the second is to realise an easy system to be maintained. In this paper, network-based subsumption architecture is proposed for a multiple mobile robots system in order to cope with the earlier mentioned two requirements. Each of multiple mobile robots has three wheels driven by DC motors and six position sensitive detector (PSD) sensors. A server supervisory control system using the networked-based subsumption architecture is designed to realise a schooling behaviour by using only information from the PSD sensors. The server supervisory control is introduced for poor hardware platforms, i.e., mobile robots that can only behave based on several simple agents according to sensing and action. The multiple mobile robots system can be used as a unique mechatronics educational system for students or engineers in the field to efficiently enable learning the basic concept of subsumption architecture to develop swarm intelligence. Experimental results show interesting behaviour among the multiple mobile robots, such as following, avoidance and schooling. The ease of maintenance, system extensibility, and educational effectiveness of the proposed multiple mobile robots system are presented.

2

Three wheeled mobile robot with six PSD sensors

2.1 Basic hardware Figure 1 shows the developed mobile robot equipped with six distance sensors (Nagata et al., 2011; Kitahara et al., 2012). The main body of the robot is an omni-directional mobile robot with three wheels provided by TOSADENSHI LTD. A MicroConverter ADuC814ARU provided by ANALOG DEVIVES is mounted on the control board of the mobile robot. A simple DC motor without encoder is built in each wheel, so that the robot has a high cost performance. The maximum moving speed and rotation speed are about 25 cm/s and 2π rad/s, respectively. Table 1 tabulates the basic specification of the mobile robot. In order to measure the distances to objects in real time, each robot is equipped with six PSD sensors as shown in Figure 2. di(k) = [di1(k),…,di6(k)]T is the distance vector of the ith mobile robot at the discrete time k. The PSD sensor is mainly composed of LEDs, electrical resistances and photodiodes, and can calculate the distance to an object through triangulation technique. In order to cope with the problem of narrow directivity, the required number of the PSD sensor was set to six. Actually, more than six PSD sensors are desirable to further reduce the dead angle. As can be seen from Figure 1, it is not easy for the robot to have as many as more than 12 PSD sensors in consideration of

Multiple mobile robots system with network-based subsumption architecture

61

the both sizes of a mobile robot and a PSD sensor. When PSD sensors fixed on a mobile robot are increased from 6 to 12, the dead angle reduces to about 30 degrees. Figure 3 shows the static measurement result of a PSD sensor, which represents the relation between the actual physical distance and the digital value measured through the PSD sensor. As can be seen, the measured values tend to vary with the rise of distance, and the sensible distance, i.e., the available range of distance, is about from 8 cm to 90 cm. The dispersion is caused by the room brightness condition, the object’s shape and the reflection condition of an LED light. The AD converter used in the mobile robot can generate digital values with 8 bits, i.e., 256 steps. The digital value is linearly tuned to be almost the same value with the actual distance as shown in Figure 3, so that the digital value in vertical axis can be regarded as the actual distance. Also, the heights of six PSD sensors are 5 cm from the floor, so that the available height for object detection is about 5 cm. Besides, each robot has a Bluetooth wireless device to communicate with a PC server. The hardware architecture and mechanism of the mobile robot are very simple. The expendable part of the robot is only three small DC motors because of the abrasion of the brushes. This robot is used to demonstrate the emergence of schooling behaviour. Figure 1

Mobile robot with three wheels and six PSD sensors (see online version for colours) RY

PSD sensor 2

PSD sensor 1

PSD sensor 3

RX

PSD sensor 6

PSD sensor 4

PSD sensor 5

Table 1

Basic specification of the three-wheeled mobile robot used for experiments

Dimensions [mm] Weight Withstand load Buttery run time Type of wheel Moving velocity

Depth 220, width 220, height 100 660 [g] 1,000 [g] 140 minutes Omini-type wheel 25 [cm/s] in case of a low gear

Gradability

10 [deg] in case of a high friction

CPU

MicroConverter ADuC814ARU

C-compiler Standard interface

KEIL EK51V720.EXE USB

62

F. Nagata et al.

Figure 2

Configuration of six PSD sensors fixed around a three-wheeled mobile robot, in which di(k) = [di1(k),…,di6(k)]T is the distance vector of the ith mobile robot at the discrete time k R

Wheel 2 di3(k)

Y

di2(k)

Wheel 1 di1(k)

PSD 2 PSD 3

PSD 1 R

O

PSD4

X

PSD6

di4(k)

Wheel 3

Figure 3

di6(k)

PSD 5

di5(k)

Relation between actual distance and digital value measured from a PSD sensor (see online version for colours)

Digital value from a PSD sensor

80 70

Max. values

60

Min. values

50 40 30 20 10 0

0

10

20

30

40

50

60

70

80

90

Real distance cm

2.2 Kinematic control of the three-wheeled mobile robot Figure 4 illustrates the kinematic model of the mobile robot in robot coordinate (O − R X RY ). ωi (i = 1, 2, 3) is the angular velocity of each wheel. Also, system

∑

R

vi (i = 1, 2, 3) is the forward velocity of each wheel and it is given by vi = rωi

(i = 1, 2, 3)

(1)

Multiple mobile robots system with network-based subsumption architecture

63

where r is the radius of each wheel. If the position and orientation vector of the robot is given by xr = [xr yr φr]T, then the velocity quantity can be represented by x r = [ xr y r φr ]T . The following kinematic equations are obtained from Figure 4 (Watanabe et al., 1998). 1 3 v1 = − xr + y r + Lφr 2 2

(2)

1 3 v2 = − xr − y r + Lφr 2 2

(3)

v3 = xr + Lφr

(4)

where L is the distance between the centre O of the robot and the centre of each wheel. Equations (2), (3) and (4) lead to the kinematic relation given by ⎛ 1 ⎜− ⎜ 2 ω ⎛ 1⎞ ⎜ 1 1 ⎜ ⎟ ⎜ ω2 ⎟ = r ⎜ − 2 ⎜ ⎜ω ⎟ ⎜ 1 ⎝ 3⎠ ⎝

3 2 3 − 2 0

⎞ L⎟ ⎟ ⎛ xr ⎞ ⎟⎜ ⎟ L ⎟ ⎜ y r ⎟ ⎟ ⎜ φ ⎟ L ⎟⎠ ⎝ r ⎠

(5)

By using equation (5), the robot can be controlled kinematically, i.e., desired behaviour designed by x r = [ xr y r φr ]T can be performed by making three wheels rotate with the angular velocity vector ω = [ω1 ω2 ω3]T. As special cases, Table 2 shows the velocity components to move to the direction of each PSD sensor. When designing the schooling mode using multiple mobile robots, six basic velocities tabulated in Table 2 are used. The important point is the direction of velocity that a mobile robot generates in ΣR, and it depends on the ratio xri : y ri . The velocity norm || x r || can be arbitrarily changed as αω = [αω1 αω2 αω3]T with a scalar α. Figure 4

Kinematics of the mobile robot with three wheels

y r R

r ω Y v1 Wheel

φr

Wheel 2

Wheel 1 R

O

v2

L

Wheel 3

v3

X

x r

64

F. Nagata et al.

Table 2

Velocity components to move to the direction of each PSD sensor, in which i (i = 1,…,6) denotes the number of PSD sensor

i

1

2

3

4

xri y ri

3

0

− 3

− 3

0

3

1

2

1

–1

–2

–1

3

5

6

Network-based multiple mobile robot system

3.1 Software development environment and its limitation The software development environment of each mobile robot uses free C language and has two limitations due to the cost reduction. The first is, the flush ROM of the mobile robot is only 8 kB, so that it is difficult to construct high level software architecture in the mobile robot. The second is, the mathematical standard library such as ‘exp ( )’ cannot be compiled. Thus, for example, it is impossible to directly apply the potential field method for path planning. In order to cope with such limitations of the software development environment, a server supervisory control scheme is considered. In the server control mode, mobile robots behave according to commands transmitted from the server. Each mobile robot integrates all information measured by six PSD sensors and transmits it to the server once, and then the server can return a simple action command to the robot based on the sensor information while considering the overall behaviour of a swarm. By means of the proposed server supervisory control, for example, the potential field method is available on the server side where the Windows Visual Studio runs as an environment for the software development, even though the mobile robot has only a poor software platform. This can be also applied to a simulation experiment of complex swarm intelligence where a software development with comparatively large scale is required.

3.2 Server supervisory control based on subsumption architecture As mentioned in the previous subsection, the proposed server supervisory control is designed for multiple mobile robots each has three wheels and six PSD sensors. For example, this system is used to support study needs of fourth year students, such as, to learn the subsumption control architecture for schooling behaviour. The subsumption control architecture is called the behaviour-based artificial intelligence, which was first proposed by Brooks (1986). Students can practically know the basic concept and effectiveness of subsumption control architecture which provides a method for structuring reactive systems from lower level to higher level using layered sets of rules, i.e., reactive behaviours according to the change of robot’s environment. The details will be introduced in the latter subsections. In this subsection, a network-based multiple mobile robots system using the server supervisory control is proposed in order to be able to design high-level software architecture in spite of the limited capability of poor robot software platform. Figure 5 shows the conceptual diagram of the network-based multiple mobile robots system where the supervisory server can interact with multiple mobile robots. Each robot can only transmit six PSD sensory information di(k) = [di1(k),…,di6(k)]T to the supervisory server through Bluetooth communication. The subscript i denotes the identification number of a

Multiple mobile robots system with network-based subsumption architecture

65

robot. The supervisory server returns a set of a simple behaviour and a short execution time, e.g., 200 ms to the corresponding robot. Eight types of the most simple and subdivided reaction behaviours, i.e., reflex actions, are prepared for the mobile robots as tabulated in Table 3. When a set of a command code and an execution time is transmitted from the supervisory server to a mobile robot, the mobile robot conducts the motion exactly within the specified execution time. Three agents called ‘avoid objects’, ‘turn to left or right’ and ‘move forward’ are designed by using the basic acts shown in Table 3 in order to realise a schooling behaviour. Figure 5

Network-based multiple mobile robots system aiming at the emergence of a schooling behaviour, where the supervisory server can interact with multiple mobile robots Supervisory Server 8 7

Bluetooth 2

4

５

2

P mm and SD se ns cod e a or inf o nd exe rmati o cut ion n tim e

3

3

9

Co

6

10

1

Table 3

Cmd. code

The most simply subdivided reaction behaviours, i.e., reflex actions, for mobile robots

Corresponding reflex action

1

Move to the direction of PSD sensor 1

2

Move to the direction of PSD sensor 2

3

Move to the direction of PSD sensor 3

4

Move to the direction of PSD sensor 4

5

Move to the direction of PSD sensor 5

6

Move to the direction of PSD sensor 6

7

Rotate to clockwise direction

8

Rotate to counterclockwise direction

4

Subsumption architecture-based controller implemented in supervisory server

It is difficult for the conventional subsumption architecture to be directly implemented in this type of low-cost mobile robot whose development environment on hardware and software is limited due to cost reduction. Our proposed network-based subsumption architecture is developed for such a mobile robot platform.

66

F. Nagata et al.

4.1 Basic design The software development environment is C# of Windows Visual Studio, which is used to develop and implement high level software architecture such as subsumption control architecture according to application requirements. Figure 6 shows the subsumption control architecture implemented on the supervisory server. The controller includes the three agents ‘avoid objects’, ‘turn to left or right’ and ‘move forward’ for a schooling behaviour of multiple mobile robots. The upper agent has a higher priority to be dispatched. This section introduces the three kinds of agents and the corresponding output command codes shown in Table 3. The eight commands shown in Table 3 are simple and basic motions while it represents important reflex actions for each mobile robot to consequently produce the competences of the three agents. Figure 6

Subsumption control architecture for a schooling behaviour, which is implemented on supervisory server PC PSD 1 PSD 2 PSD 3 PSD 4 PSD 5 PSD 6

Avoid objects

Virtual wheels

Wheel 1

Turn to left or right Move forward

Wheel 2 S

S

Wheel 3

Subsumption architecture on supervisory server PSD sensor information

Mobile robot 1

Command code & execution time

. . .

PSD sensor information

Mobile robot N

Bluetooth communication

Bluetooth communication

Virtual PSD sensors

Command code & execution time

The server receives PSD sensors information di(k) (1 ≤ i ≤ N) from all mobile robots every sampling period of 100 ms, in which N is the number of the available mobile robots within the system. By analysing di(k), the controller dispatches the current execution right to one of the three agents for the ith mobile robot. This process is periodically applied to all mobile robots in the order from 1 to N. In the schooling mode, all mobile robots try to regularly move along the inner of a circular fence keeping the distance to both the fence and other mobile robots. This mode enables the robots to behave like carps in a Japanese artificial circular pond.

4.2 ‘Move forward’ When a robot moves counterclockwise along the inside of the circular fence as shown in Figure 5, the agent of ‘move forward’ is always active. Accordingly, if higher priority agents such as ‘turn to left or right’ and ‘avoid objects’ are inactive, then the agent of ‘move forward’ can have the execution right and the following control law is applied. x i = v

x i 2 x i 2

(6)

Multiple mobile robots system with network-based subsumption architecture

67

where x i = [ xi yi ]T is the virtual translational velocity for the ith mobile robot,

x i 2 = [0 1]T is the normalised velocity vector for moving to the PSD sensor 2. The position of PSD sensor 2 is assumed to be the front of the mobile robot. v is the scalar signifying the magnitude of the velocity. While the agent of ‘move forward’ has the execution right, a set of a command code 2 and an execute time T, e.g., 200 ms, are continuously transmitted to the mobile robot every the specified execution time T.

4.3 ‘Turn to left or right’ The orientation of a mobile robot is controlled by the agent of ‘turn to left or right’. This agent becomes active when di1(k) < dref and di6(k) < dref are simultaneously satisfied. dref is called the active reference range. Further, if this agent has the execution right, then the following orientation control law is applied.

φi = Kφ {di 6 (k ) − di1 (k )}

(7)

where φi is the virtual rotational velocity of ith mobile robot, di1(k) and di6(k) are the values of PSD sensors 1 and 6 transmitted from the ith mobile robot, respectively. As can be seen from Figure 2, PSD sensors 1 and 6 are selected for the orientation control moving counterclockwisely. Kφ is the gain that can control the orientation of the robot to be parallel to the inner side of the circular fence. In this case, a set of a command code 7 or 8 and an execution time T are continuously transmitted to the ith mobile robot every specified execution time. The command code 7 or 8 is determined by the sign of φi . The agent ‘turn to left or right’ has a higher priority than ‘move forward’. Thus, while the agent of ‘turn to left or right’ has the execution right, the agent of ‘move forward’ is suppressed, i.e., blocked.

4.4 ‘Avoid objects’ The agent of ‘avoid objects’ has the highest priority. If ∃dij(k) [dij(k) < dd] becomes true, then this agent becomes active, i.e., has the execution right at the same time. After becoming active, this agent generates the virtual velocity given by x i = −v

x ij x ij

{d

d

}

− min dij (k ) j

{∃dij (k ) [ dij (k ) < dd ]}

(8)

where dd is the minimum allowable distance between a robot and an object. The supervisory server transmits a set of a command codes j + 3 (in case of j = 1, 2 or 3) or j – 3 (in case of j = 4, 5 or 6) and an execution time T to a mobile robot in order to avoid a collision with the nearest robot or object. Due to the activation of ‘avoid objects’ agent, the mobile robot can move away from moving objects including other mobile robots within the collision critical zone. If the distance to the nearest object is smaller than dd, the robot tries to expand the distance to be over dd. When multiple PSD sensors simultaneously detect shorter distances than dd, the mobile robot tries to preferentially move away from the nearest object. Figure 7 shows an experimental scene of schooling behaviour, in which multiple mobile robots are controlled based on the subsumption control architecture incorporated

68

F. Nagata et al.

in the supervisory server as shown in Figure 6. It was confirmed from the experiment that the multiple mobile robots could perform a desirable schooling behaviour. Figure 7

Experiment of schooling behaviour based on the proposed networked-based subsumption architecture (see online version for colours)

4.5 Agent dispatcher The agent dispatcher supervises the controller shown in Figure 6. The dispatcher does not immediately move the execution right to a higher priority agent when a lower priority agent is running, because each agent works as a simple reactive behaviour every sampling period according to sensor information. Instead of this, whenever a reaction behaviour is executed during a specified execution time, the dispatcher checks and updates the activity of each agent and gives the next execution right to a newly updated active agent with the highest priority.

4.6 Multiple sensory sensors The mobile robot introduced in this paper basically has six AD conversion channels, however, all of which are already connected to six PSD sensors. In order to be able to deal with sensory information from other sensors such as temperature, humidity, force and smell, a compact AD conversion module AGB65-ADC provided by Asakusagiken Co., Ltd. is mounted on the mobile robot. The AGB65-ADC is a high performance AD converter that can further handle 16 analogue channels. Allowed analogue input range is from 0 to 5 V, and an analogue value is converted to either of two types of digital values, i.e., the maximum is 255 (8 bits) or 4,095 (12 bits). Figure 8 illustrates the hardware block diagram showing the connection scheme in a mobile robot including analogue sensors, a AGB65-ADC module and a Bluetooth module called Bluemaster. The mobile robot has one serial port for data communication, so that text codes can be transmitted to and received from the Bluetooth module through AGB65-BT to the serial port 2 in AGB65-ADC. The Bluetooth module is used to communicate with the supervisory server PC. It should be noted that when communicating with the AD converter in AGB65-ADC to obtain sensor’s information through serial port 1, command sequence including a hexadecimal header code ‘0xff’ must be transmitted. In other words, the redirectable serial port in AGB65-ADC board can switch data to serial port 1

Multiple mobile robots system with network-based subsumption architecture

69

or to serial port 2 automatically, which is the important redirect function of AGB65-ADC. Using the AD conversion module AGB65-ADC, students can easily further add various types of small and light weight sensory sensors as shown in Figure 9 to a mobile robot with six PSD sensors. Figure 8

Hardware block diagram of a mobile robot with a compact AD conversion module AGB65-ADC and a Bluetooth module Bluemaster (see online version for colours) To server PC

Mobile robot Bluemaster

16 analog sensors …

AGB65-BT A/D converter Serial port 2

Serial port 1

Redirectable serial port AGB65-ADC Serial port MicroConverter (ADuC814ARU)

Figure 9

Various types of small and light weight sensory sensors provided by Asakusagiken Co., Ltd., Japan (see online version for colours)

Smell (gas) sensor

Photo reflector

Film type force sensor

Temperature sensor, Humidity sensor

Flex sensor

Protrusion type force sensor

70

5

F. Nagata et al.

Conclusions

In this paper, a unique multiple mobile robots system has been proposed and implemented for students pursuing engineering studies in the field to experimentally learn the basic concept of subsumption control architecture that reflects a typical behaviour-based artificial intelligence. Each of the multiple mobile robots has three wheels and six PSD sensors, where each wheel is driven by the most inexpensive type of DC motor with no encoder. Network-based subsumption architecture has been presented to realise a schooling behaviour by using only information from the PSD sensors. The proposed server supervisory control is designed for poor software and hardware platforms, i.e., mobile robots that can only behave based on simple agents according to sensing and action behaviours using the PSD sensors. Experimental results showed interesting behaviour among the multiple mobile robots, such as following, avoidance, and schooling. The educational effectiveness of the proposed multiple mobile robots system was confirmed through experimental instructions at Tokyo University of Science, Yamaguchi, Japan. It is expected that the proposed system will be able to be used for a basic experimental system for swarm intelligence using multiple mobile robots.

References Antonelli, G., Arrichiello, F. and Chiaverini, S. (2009) ‘Experiments of formation control with multirobot systems using the null-space-based behavioral control’, IEEE Transactions on Control Systems Technology, Vol. 17, No. 5, pp.1173–1182. Antonelli, G., Arrichiello, F. and Chiaverini, S. (2010) ‘Flocking for multi-robot systems via the null-space-based behavioral control’, Swarm Intelligence, Vol. 4, No. 1, pp.37–56. Antonelli, G., Arrichiello, F., Chakraborti, S. and Chiaverini, S. (2007) ‘Experiences of formation control of multi-robot systems with the null-space-based behavioral control’, Procs. IEEE International Conference on Robotics and Automation, April, Roma, pp.1068–1073. Azarm, K. and Schmidt, G. (1997) ‘Conflict-free motion of multiple mobile robots based on decentralized motion planning and negotiation’, Procs. IEEE International Conference on Robotics and Automation, April, Albuquerque, NM, USA, pp.3526–3533. Barman, A.R., Kingdon, S.J., Little, J.J., Mackworth, A.K., Pai, D.K., Sahota, M.K., Wilkinson, H. and Zhang, Y. (1993) ‘DYNAMO: real-time experiments with multiple mobile robots’, Intelligent Vehicles Symposium, July, Tokyo, Japan, pp.261–266. Bennewitz, M. and Burgard, W. (2000) ‘A probabilistic method for planning collision-free trajectories of multiple mobile robots’, Procs. of the workshop Service Robotics – Applications and Safety Issues in an Emerging Market at the 14th European Conference on Artificial Intelligence, August, Berlin, Germany, 7p. Brooks, R.A. (1986) ‘A robust layered control system for a mobile robot’, IEEE Journal of Robotics and Automation, Vol. 2, No. 1, pp.14–23. Guo, Y. and Parker, L.E. (2002) ‘A distributed and optimal motion planning approach for multiple mobile robots’, Procs. IEEE International Conference on Robotics and Automation, Washington, DC, USA, pp.2612–2619. Kitahara, N., Nagata, F., Otsuka, A., Sakakibara, K., Watanabe, K. and Habib, M.K. (2012) ‘A proposal of experimental education system of mechatronics’, Procs. of 17th International Symposium on Artificial Life and Robotics, January, Beppu, Oita, Japan, pp.166–169. Kube, C.R. and Zhang, H. (1992) ‘Collective robotic intelligence’, Procs. the Second International Workshop on Simulation of Adaptive Behavior, pp.460–468. Kube, C.R. and Zhang, H. (1993) ‘Collective robotics: from social insects to robots’, Adaptive Behavior, Vol. 2, No. 2, pp.189–219.

Multiple mobile robots system with network-based subsumption architecture

71

Nagata, F., Yamashiro, T. and Watanabe, K. (2011) ‘Cooperative swarm control for multiple mobile robots using only information from PSD sensors’, Artificial Life and Robotics, Vol. 16, No. 1, pp.116–120. Noreils, F.R. (1992) ‘An architecture for cooperative and autonomous mobile robots’, Procs. IEEE International Conference on Robotics and Automation, May, Nice, France, pp.2703–2710. Parker, L.E. (2003) ‘The effect of heterogeneity in teams of 100+ mobile robots’, Multi-Robot Systems: From Swarms to Intelligent Automata, Vol. 2, pp.205–215, Kluwer Academic Publishers. Parsons, D. and Canny, J. (1990) ‘A motion planner for multiple mobile robots’, Procs. IEEE International Conference on Robotics and Automation, May, Cincinnati, OH, USA, pp.8–13. Pimentel, B. and Campos, M. (2003) ‘Cooperative communication in ad hoc networked mobile robots’, Procs. IEEE/RSJ International Conference on Intelligent Robots and Systems, October, Las Vegas, NE, pp.2876–2881. Watanabe, K., Shiraishi, Y., Tzafestas, S.G., Tang, J. and Fukuda, T. (1998) ‘Feedback control of an omni-directional autonomous platform for mobile service robots’, Journal of Intelligent & Robotic Systems, Vol. 22, Nos. 3/4, pp.315–330.