A study of maze searching with multiple robots system

Shinichiro Yano
Kawasaki Heavy Industries, LTD., 1 Kawasaki-cho, Kakamigahara City, Gifu 504 Japan
Manabu Noda, Hisahiro Itani, Masayuki Natsume, Haruhiko Itoh and Hajime Hattori
Nagoya Municipal Industrial Research Institute, 3-4-4 1 Rokuban, Atsuta-ku, Nagoya 456 Japan
Tadashi Odashima, Kazuo Kaya, Shinya Kataoka and Hide0 Yuasa
Graduated School of Engineering Nagoya University, Furo-cho, Chikusa-ku, Nagoya 464-01 Japan
Xiangjun Li
Department of Mechanical Engineering, Daido Institute of Technology
2-2 1 Daido-cho, Minami-ku, Nagoya 457 Japan
Mitsuhiro Ando, Wataru Nogimori and Takahiro Yamada
AISIN COSMOS R&D CO., LTD., 5-50 Hachiken-cho, Kariya, Aich 448 Japan

Abstract: This paper shows a system composed of multiple mission is not monotonous, but some parts of it are rather
robots which cooperate each other through local easier and they can be executed autonomously. For example,
communications, in order to search a goal of a given maze. let’s thmk of repairing unknown troubles occurred in a
Though a diameter of an each robot is 40 mm and it equips sewerage system. In t h s case, it is difficult to find the
with a very simple algorithm, each robot plays a part to find position and the cause of the trouble and repair it by fully
the goal of an unknown maze as a whole system. In this study, a autonomous agents, but just findmg the position sometimes
computer simulator has been also developed in order to much easier for autonomous agents, though the sewerage
examine the validity of designed algorithms.
system is very huge one.
From this point of view, we have studied maze searching.
1. INTRODUCTION In this study, our goal is to develop a demonstration system’
to search a goal of a given maze, though the location of a
In these days, autonomous systems have been used in goal and the structure of a given maze is unknown. We have
the various fields and are expected even in the fields which constructed the actual multi-robots system and have
a help of human beings has been indispensable such as demonstrated cooperative motion of them[3] [4][5]. We
patrolling, checking and repairing some troubles. For revised our previous robots for this study.
example, let us t h d about a maintenance of some social In this paper, specifications of the developed system are
system. As the scale of such contemporary systems becomes shown in section 2. A motion algorithm of each robots is
lager and more complicated, the cost to keep it only by described in section 3. In section 4,we show a simulator,
human beings tends to be very high. In the case to automate which is developed to investigate efficient algorithm of
such tasks for reducing the cost by applying robots, they robots. Next, we evaluate the obtained result and discuss its
must be multiple robots system in order to cover wide area usefulness. At last, we conclude this paper.
sufficiently. And there is no doubt that autonomous system
is desired for such kinds of tasks. 2. SYSTEM DESIGN
Autonomous agents are mobile versatile machines
capable of interacting coherently with an environment and
executing a variety of tasks in unpredictable conditions[ I]. 2.1 System Overview
Autonomy means capability of navigating the environment; This system is composed of two items:
navigation, in turn,necessarily relies on a topological and 0 Maze,
metric description of the environment[2]. 0 Robots.
Autonomous systems have been required and studied for Each item is very simple as described as follows. T h s
long time, and it is widely understood that full autonomy in means that the architecture of the proposed system is
unpredictable condtions are very difficult under some
philosophcal problems so called “Frame Problem”. But at
the same time, we also know that a difficulty of one ’ It will be demonstrated during MHS’96

Seventh International Symposium on

Micro Machine and Human Science
0-7803-3596-1/96$4.0001996 IEEE. 207
applicable to many kinds of exploration problems in real <c5> Walls on aisles are movable and make various maze
fields. patterns (refer Fig. 4).
At first, there are some number of robots at a starting There is no closed loop2in the maze.
point. Then each robot starts in order to search a goal found
in a given maze, according to its motion algorithm. No robot Under these conditions, the maze are divided into the
knows where the goal is and how the topology of the given four poidareas shown in Table 1 by Crossing line.
maze is. In this study, the global communication among
robots are not supposed. Therefore, each robot can

t
able 1. posihons of robot on the maze
communicate and cooperate another one only when they
Crossing(on) A point where robot turns its direction.
meet in the maze. The aim of robots are just finding the goal
(only one at each crossing)
set somewhere in the maze. We have supposed that all of
Crossing(in) From detecting a Crossing line until
the robots don’t have to reach the goal, because in some
Crossing(on)
case of micro machine systems, it can be assumed that the
cost of each robot is not so expensive and negligible. Crossing(0ut) From Crossing(on) until detecting a
Crossing line
I Aisle except above
2.2 Maze Design
The maze whch is assumed here is 2-dimensional, but it
can be extended to 3-dimensional one at any time if 2.3 Robots Design
necessary, as mentioned in section 5. The maze is composed About 20 robots are to be prepared for our
of following objects. demonstration. All of these robots are homogeneous and the
0 Aisle design concept of the robot is based on the previous
0 Crossing study[5]. Table 2 and Fig.2 show the specifications and
0 Wall outlook of the robot, respectively.
0 Start and Goal points
Table 2 Specificabons of the robot
In order to make the system simple, there are some
restrictions in this maze, described as below. -
Size diameter 40”
<c1> The width of the aisle is almost same of the robot height 32”
diameter. Moving mechanism Stepping Motor (x2)
<c2> A line which is detectable by a robot‘s sensor is Max. speed 5 0 d s e c along the straight way
Battery Silver-Oxide 6 2 V (1 55 V x4)
drawn on the aisle of the maze. (We call this line “Trace
Sensors
line” from now on.)(refer Fig. 1) Line sensors Photo-reflectors
<c3> There are other detectable lines at the entrance of the Touch sensors Hand made mechanical sensors
crossing. (We call this line “Crossingline” &om now on.) -
CPU
<C4> All aisles are at right angles at crossings and number Size 19 mm x 19 mm x 10 mm *
of corner is four. Package 64-pin quad flat package
YO 46 general purpose InputlOutput
Memory 16 Kbytes ROM, 5 12 bytes EPROM
and 5 12 bytes RAM
Power consumution - 600mW

* mcludmg the socket

With these specifications, the following functions are

realized.
<F1> Go forwardhackward, turn and stop by dnving two
stepping motors.
<F2> Go along Trace line with detecting it by its sensor.
<F3> Detect which section of the maze (dvided by
Crossing line) it is.

A topology of a maze, that there IS a probability to reach the visited

Fig. 1 An example of positiodareas of maze (robot is moving from left to down) crossmg wthout turrung and reachmg a goal If there is no closed loop in a
maze. it can be mterpreted as a dlreeted tree

208
<F4> Detect a physical contact at the position of touch robots on the search-tree (based on Left-hand rule).
sensors (forwardhackward).
c)Others
<F5> Communicate by 4bit/unit with another robot whch it
Left-hand Rule: A method of depth-first search in a
meets on the maze.
<F6> Count a distance between any two points of time. maze. A robot with t h s rule turns left at every crossing.
<F7> Decide its action accordmg to its motion algorithm
with information gotten from the sensors and through 3.2 Features
communications. In this study, we assumed following conditions in order
to keep it practicable.
Simdicitv: An applied algorithm must be simple and
BOTTOM BOARD UPPER BOARD use few memory to load on a small machine which has
severe restrictions on installing space and available
electric power.
Local communication: Global communication among
robots is not assumed in order to make it effective under
various circumstances.
Low dependencv on environments: The algorithm does
not depend on a topology of a maze and primary
LSIs for remote control knowledge about it. The algorithm can work even if the
lnfrared LED
batteries structure of a maze is changed dynamically.

3.3 The CDAC algorithm

In this section, we show the motion algorithm of the
maze searching robots in this system, called CDAC (Close
Dead Alley by a Collision).

connector plns 3.3.1 Outline of the algorithm

The state transition map is shown in Fig.3. The stay state
Fig2 Structure of the robot and other states which are easy to add the transition map
and special function of the last robot are ignored to make
h s map simple.
Each robot begins to move forward from start point
3. MOTION ALGORITHM with line tracing (S 1).
When a robot detects the crossing line (S2), it goes
into the center of the crossing (S3), turns to the left
3.1 Terminology (S4), and goes out of the crossing (Sl).
In this section, the terms which are used to describe the When the front touch sensor of a robot is on in the
algorithm are defined. aisle, it means that there is a wall or a retum robot. The
a) Direction of Robots: the direction of a robot on the maze aisle is dead alley and the state of the robot becomes
is divided into following two. return (Rl). The robot turns over and goes back along
SEARCH: a direction of a robot which is going forward the aisle (E).
to search a goal. When the search state robot (S4) detects somethmg by
RETURN: a direction of a robot which is going back on the front touch sensor at the crossing (out) (S5), it goes
the known aisle after it came across a dead alley. backward to the crossing (on) (S6). If there are
b)A Name of Robots: a given name of robots which is in unsearchmg aisles in the crossing, the robot turns to
some specific status. the right and begins to search another aisle (S4). If
WALL robot: A robot whch stops at the entrance of a there are no unsearching aisles, the robot becomes
dead alley and does not let other searching robots go. return state (R5) and goes out the crossing (R2).
STAY robot: A robot which is stopping because there is When the return state robot (R2) detects the crossing
another robot forwarclhackward. line (R3), it goes into the center of the crossing (R4).If
LAST robot: A robot which is located on the last of all there are unsearching aisles in the crossing, the robot

209
 [detect Crosslng Ime]

Fig.3. State transition map of the CDAC

turns 90 degrees to the left and becomes search state In the algorithm CDAC, communication between robots
(S4). If there are no unsearching aisles, the robot also is only used to find out whether the obstacle it collides is
turns 90 degrees to the left (R5) and returns out the another robot or not. We can extend functions of the robot
crossing (R2). further, by utilizing communication. For example, if the
When the return state robot (R2)detects somedung by robots exchange information about the topology of the maze
the front touch sensor (R6), it must collide a search which they have searched, then they can achieve their goal
state robot. They both go back to the nearest crossing much more efficiently.
and if there are unsearching aisles in the crossing, the
return state robot becomes wall robot (R7).If there are 4. SIMULATOR
no unsearching aisles, it means that the crossing is on
the way of the dead alley. The robot turns on the "stop We have also developed a simulator to design motion
at the next crossing" flag, goes into the crossing (R3), algorithms of robots easily bnd effectively. The outlook of
turns 90 degrees to the left (R4, R8), and returns out this is shown as Fig.4.
the crossing (R6). This is developed by over Microsoft Visual C++ Ver.4.0,
and it is executed on MS-Windows 95NT3.51. It realize a
asynchronous parallel discrete event simulation (PDES)[6],
3.3.2 Communication
Robots can communicate each other when they meet on and this is a test bed to develop and test motion algorithms.
the maze. With our specifications of robots, just four bits
are available for one exchange. 5. EVALUATION AND CONSIDERATION

210
 Even if only one robot become a wall robot on the right
way by a mistake, it is serious accident for the system. Since
huge number of collisions and communication between
robots are expected, the probability of the occurrence of the
above accident is not low. It will be necessary to give the
robot some functions that even if it become wall robot, it
restore automatically after some collision-less periods.

6. CONCLUSION

In this paper, a maze searching system which is prepared

a ............. ~ I . . . . - . - -.-...--..-.--’
VI 1 I to demonstrate an example of a micro machine system is
presented. The main point of the proposed system is to
demonstrate that a simple algorithm over lower functions
which can be loaded into micro machines works well to
realize autonomous searching motions. Even if all of micro
machines don’t show expected motions, with the advantage
Goal
of lots of number of them, we can find that they accomplish
desirable missions as a whole system.

7. ACKNOWLEDGMENTS
Fig.4. An outlook of the maze simulator. Dashed line shows right
We are grateful to Prof. T. Fukuda of Nagoya University
way of the maze. Agents in the circles are wall robots and a
for some useful discussions. We are also thankful Mr. K.
return state robot is indicated “R”. Others are search state.
Tominaga of Nagoya City Hall, and Dr. M. Tsutsui and Mr.
K. Itoh of Nagoya Urban Industries Promotion Corporation
for their help.
5.1 Evaluation
We adopted a left-hand rule which is one of the depth- REFERENCES
first search methods, but it is not difficult to change the
search methods such as width-first search since the program [11 D. Nitzan, Development of intelligent robots:
of the simulator consists of sub-routines and the ROM of achievements and issues, IEEE Journal of Robotics and
the CPU of the robot is rew-ritable. Automation RA-l(l), pp.3-13, 1985
Since we want to make the robot as small as possible, [2] D. Mario, and S. Rizzi, Unsupervised Multi-Agent
the simulator is useful for choice the minimum equipment of Exploration of Structured Environments, Proc. of First Int.
the robot. The simulator is also used for the check of the Conf. of Multi-Agent Systems, pp.269-275, 1995
actual parameter such as the time interval of the stay state [3] M. Asano et al., Prototype of Distributed and
and the distance of the backward motion of robots for Cooperative System of Autonomous Micro Mobile Robots,
optimization of the whole system. 4th Int. Symp. on Micro Machine and Human Science,
The simulator also shows clearly that many robots have Nagoya, pp. 134-140,1993
to stop on the aisles and become wall robots. This means [4] M. Noda et al., Distributed and Cooperative System of
that efficiency of the system is not so high. Therefor, the Micro Robots with Command Delivery, 5th Int. Symp. on
last robot was considered to improves such situation. The Micro Machine and Human Science, Nagoya, pp. 111-114,
robot can transfer the last flag to another one it contacts first 1994
and restores the contacted robot after the robot passed by. [SI H. Itani et al., A Study of Mini-Robots Bucket Brigade
The new last robot do the same for the nearest wall robot, System, 6th Int. Symp. on Micro Machine and Human
and all wall robots restore one by one. Science, Nagoya, pp.205-209, 1995
[6] Y. Lin, and P. A. Fishwick, Asynchronous Parallel
Discrete Event Simulation, IEEE Trans. On Systems, Man,
5.2 Application and remained problems Cybernetics (A) Vo1.26, No.4, pp.397-412, 1996
The algorithm which we adopted is not restricted by the
dimension, this study is easily extend to 3-dimensional maze
if robots can move vertical direction.

211