223

Vol. 15 No. 6 December 2003

DEVELOPMENT OF AGENT-BASED AUTONOMOUS ROBOTIC WHEELCHAIR CONTROL SYSTEMS

CHUNG-HSIEN KUO, HSU-LUNG HUANG, MrNG-YiH LEE

Department of Mechanical Engineering, Chang Gung University, Tao-Yuan, Taiwan

ABSTRACT
In this paper, the intelligent and agent-based autonomous robotic wheelchair controller is proposed. The proposed agent-based autonomous robotic wheelchair control system consists of the functions of path planning, navigation and obstacle avoidance. In this work, the A* algorithm is used to develop the path planning module; the fuzzy logic is used to develop the obstacle avoidance and comfortable wheelchair motion control. Especially, the fuzzy fusion method is used to integrate the proposed fuzzy agent model. Finally, several cases of the individual agent execution, fuzzy module fusion, path planning, and wheelchair coordination are discussed, and the simulation results show that the proposed autonomous robotic wheelchair controller achieves the desired objectives. In addition, the agent-based design increases the model integrability and maintability. Biomed Eng Appl Basis Comm, 2003 (December); 15: 223-234. Keywords : Multi-agent system; autonomous robotic wheelchair; fuzzy system; obstacle avoidance

1. INTRODUCTION Taiwan is being an aging society. The ration of the elder person is around 8.7 percent in 2002, and it is estimated to be 25 percent at 2040 [20]. Hence, improving the environment of care for elders is more important than ever. They promote a more feasible, comfortable and safe living environment. Wheelchair is one of the most important mobility aids for elders and disabled patients. In general, the elder person and disabled patients cannot drive the electrical wheelchair cxpeditiously. It will result in the danger of collision. In recent years, several autonomous wheelchair [2-3, 10, 12, 17] systems were developed. These works focused on improving the autonomy, intelligence, safety, and navigation of the wheelchair. The VAHM (French acronym for Autonomous Vehicle for people Received: Oct 1, 2003; Accepted: Nov 6, 2003 Correspondence: Ming-Y.h Lee, Professor Ming-Yih Professor r-> __. * r * x. Engineering, Chang Gung ,-ii_ ^ Department of Mechanicali r Department ol Wen-Hwa bngineenng, Chang TaoUniversity, 259,Mechanical 1st Road, Kwei-Shan,uung University, 259, Wen-Hwa 1st Road, Kwei-Shan, TaoYuan, Taiwan, 333 Yuan, Taiwan, 333 E-mail: ieemiy@mail.cgu.edu.tw E-mail: ieemiy@mail.cgu.edu.tw

with Motor Disabilities) [2] project developed an Motor Disabilities) project developed an autonomous wheelchair to assist the disabled people to to drive a conventional powered who are unpractical to drive a conventional powered The wheelchair. The VAHM robot software control architecture is categorized as the physical, local and as global levels. In the local level, the freespace detection, wall detection, localization implemented the w a u detection, and localization are implemented as the perception controller; the freespace search, direction following, wall following, path planning, motion control, obstacle avoidance are implemented as the as navigation controller; and the man-machine interface and o m m a n d interpretation are implemented as the a n d ccommand interpretation are implemented as the communication controller. The SENARIO (Sensor Aided Intelligent Intelligent T h e SENARIO (Sensor Aided Wheelchair Navigation System) [3] project developed an intelligent wheelchair navigation system, and it a n intelligent wheelchair navigation system, and it provides fully autonomous p r o v i d e s fully autonomous and semi-autonomous the fully autonomous mode, the control modes. In the fully autonomous mode, the control system can accept the user's command. When the o n t r 1 y s t e m accepts a "go to goal" command, it fhe ccontrol ssystem ? * "8 o goal" command, it v locates tthe positions of 1itself aand t the targeti n t eterms f in of es he c a tenvironmental, omap. l ^ n d h e ^ f P o s l t l o t n s f , s e Consequently, a rpath is o ,the ne environmental map. Consequently, following generated, and the wheelchair executes bya path is gthee rdesired apath. Inwaddition, the wheelchair following en atetj5 nd the h e e lchair executes by can avoid t h e d e s i r e d p a t h I n addition, the wheelchair can avoid obstacles in terms of the ultrasonic and infrared obstaclesIn the semi-autonomous mode, the user's sensors. in terms of the ultrasonic and infrared sensors. In the semi-autonomous mode, the user's

with

-12 -

BIOMEDICAL ENGINEERINGAPPLICATIONS, BASIS & COMMUNICATIONS

command can override the action generating from the autonomous control mode, and the users can drive the wheelchair using a specified path in a special environment. The Bremen Autonomous Wheelchair project [8] developed a "Rolland" wheelchair, and it is implemented by following the skills of the driving and route assistants. The "Rolland" is capable of smooth speed control, turning around in an open space, turning around in a comer, obstacle avoidance, and doorway passage. It provides the mobility assistance for the handicapped people. On the other hand, the SIAMO (Spanish acronym for Integral System for Assisted Mobility) project [12] developed an electronic system to guide the autonomous wheelchairs for the disabled and elder people. The proposed autonomous wheelchair system integrates an innovative usermachine interface, a complete sensory subsystem (ultrasonic, infrared, vision, etc), and an advanced strategy of control and navigation system to guide the wheelchair with safety and comfort.

224 autonomous software or hardware component. In general, the agent has the following characteristics:

These literatures pointed out the importance of the autonomous wheelchair. Therefore, this paper proposes an agent-based autonomous robotic wheelchair controller. The proposed agent-based control architecture controls the autonomous robotic wheelchairs in an indoor environment. The navigation system of the autonomous robotic wheelchair is developed as a group of intelligent agents, including the A* based robotic wheelchair path planning and the fuzzy logic based motion control and obstacle avoidance. In addition, the multiple robotic wheelchairs' coordination mechanism is also designed to handle the cooperation, management, and emergency handling of the robotic wheelchairs with safety and efficiency. Finally, several cases of the individual agent execution, fuzzy module fusion, path planning, and wheelchair coordination are discussed, and the simulation results show that the proposed autonomous robotic wheelchair controller achieves the desired objectives.

1. Autonomy: the agent can operate in a standalone environment, and it is capable of self-monitoring. 2. Socialability: the agent can communicate with another agents or users. 3. Reactivity: the agent can detect the environmental information and respond to it. 4. Proactiveness: the agent can execute the desired task. Hence, the agent can sense the environmental information and infer the facts in terms of the knowledgebase. The agent can also communicate with another agents and interact with the real word. Therefore, the agent-based system can operate in an unpredictable and dynamic environment. In addition, the distributed multi-agent coordination system [4-5, 16] is defined as the agents that share the desired tasks in a cooperative point of view, and they can be autonomously executed at different sites. Fig. 1 shows the architecture of the distributed multi-agent coordination system [16]. Note that each agent can execute at different network node. 2.1.2 Distributed agent-based robotic wheelchair control architecture Autonomy is the most important characteristics of the agent technology. Therefore, it is suitable to construct the autonomous robotic wheelchair control system. The distributed agent-based robotic wheelchair control architecture can be divided into three-layer hierarchy: the supervisory control layer; the path planning layer; and the navigation layer, as shown in Fig. 2. The detailed descriptions of this control architecture are further elaborated as follows.
The navigation layer is designed to guide the robotic wheelchair moving with safety, comfort and

2. METHODS
2.1 Distributed Agent-Based Robotic Wheelchair Control Systems
2.1.1 Distributed multi-agent coordination system The proposed distributed agent-based robotic wheelchairs controller is developed using the distributed multi-agent coordination control architecture. The agent technology [6, 18] is a new and important technique in recent novel researches in the artificial intelligence (AI) area. The agent technology is defined as a communicable and intelligent

Communication type agent

Perception Coordination Executive type agent type agent type agents _JZ!

Environment Fig. 1 Distributed multi - agent coordination architecture

-13-

225
r --------------------------------

Vol. 15 No. 6 December 2003 handling agent is executed when an abnormal signal is detected and recognized. Such events will directly inform the corresponding doctors and nurses. 4. Multi-robotic wheelchair coordination agent: The coordination mechanism of multi-robotic wheelchairs is done in terms of the predefined wheelchair priorities. When the paths of two robotic wheelchairs may intersect, the lower priority one will parry away from the path of the higher priority one. Such a mechanism ensures highest priority robotic wheelchair can reach its target as soon as possible. 5. Reporting agents: the reporting agent can be executed in terms of time-driven or eventdriven operations. It delivers information to the related doctors or nurses through the Internet. 2.2 Development of Robotic Wheelchair Navigation Systems In this work, three navigation modules [9, 15] are developed, including the wall-following, goal-seeking, and obstacle avoidance. They are developed using the fuzzy logic [1, 11, 13]. The fuzzy system uses the fuzzy set and the membership function to describe the facts with uncertainty. The inference engine is executed according to the input status and knowledgebase to determine the control policy.
The robotic wheelchair has 7 ultrasonic sensors, and they are used as the inputs of the navigation agents, as shown in Fig. 3. The operation of the navigation agents is indicated as in Fig. 4. In this figure, three modules are dynamically switching, and the switching mechanism depends on the ultrasonic signal . Such a dynamic switching mechanism forms the fuzzy fusion method. The switching reference table of the fuzzy fusion method is shown in Table 1. In this table, the ul to u7 indicate the signals collected form the ultrasonic sensors. Note that "X" indicates no obstacle detected; "0" indicate any obstacle detected; and "-" indicates "don't care". In addition, the velocity control policy can be inferred from the results of the wall-following, goal-seeking, and obstacle avoidance agents. Note that the velocity control agent module is designed to drive the wheelchair with comfort and safety.

I Supervisory control layer agent

- --- -----------------------r-------------------------------Subgoa I generefing agent 1

Navigation layer L -------------------------------'

Fig. 2 Distributed agent -based multi-robotic wheelchair control architecture efficiency. In this paper, three navigation modules are used, including the wall-following, goal-seeking, and obstacle avoidance. They are developed using the fuzzy logics. Three navigation modules are dynamically switching in terms of the navigation coordination agent in terms of the fuzzy fusion method. In addition, the steering control agent is designed to control the angular velocities of the wheels. The agents in the path-planning layer include: 1. Path planning agent: It uses the A* search algorithm [14] and the visibility graph to determine the minimum path. This agent generates a set of straight lines connecting the start point, the obstacle vertices, and the target point. All connected lines are not allowed to intersect with the obstacles. Finally, the optimal path can be determined in terms of minimizing the cost function that is calculated from the Euclidean distance. 2. Sub-goal generating agent: When the optimal path is found, the sub-goal generation agent records the set of connected straight lines. Consequently, these straight lines will be transformed to the sub-goal sequence. Finally, in order to increase the efficiency of the overall multiple autonomous robotic wheelchair system, a supervisory controller is required, and it consists of: 1. Database access agent: The database is used to store the electronic map, robotics wheelchair trajectory and location information. Note that the database access agent is developed using the open database connectivity (ODBC) [19]. 2. Remote monitoring/ control agents: It can monitor the localization and status of the robotic wheelchairs.
3. Emergency handling agent: The emergency

2.2.1 Goal-seeking mode The goal-seeking mode is defined as none of the ultrasonic sensor detecting any obstacle, and the sensors in the same side ("S4 and S5" or "S6 and S7") do not detect obstacles simulataneously (as indicated in Table 1) . In this mode, the fuzzy rule table is shown in Table 2. Where 0 is the angle between the direction of the wheelchair (a) and the direction of from two rear wheels' centre to the target (13 ), as shown in Fig. 5;,u is the steering coefficient; NB indicates negative big;

-14-

BIOMEDICAL ENGINEERINGAPPLICATIONS, BASIS & COMMUNICATIONS

226
NM indicates negative medium; NS indicate negative small; ZE indicates zero; PS indicates positive small; PM indicates positive medium; and PB indicates positive big. The input membership function of 0 and the output membership function of a are shown in Fig. 6(a) and Fig. 6(b), respectively.

750 mm

Fig. 3 Robotic wheelchair ultrasonic sensor layout and mechanical geometry

2.2.2 Obstacle avoidance mode The obstacle avoidance mode is defined as when any of the front ultrasonic sensor (Si, S2, or S3) detects the obstacle. In this mode, the fuzzy rule table is shown in Table 3. Their membership functions are shown in Fig. 7. Where Dsi is the distance measured from Si. 2.2.3 Wall-following mode The wall-following mode is defined as Si, S2, and S3 do not detect any obstacle and the two ultrasonic sensors in the same side detect obstacles. In this mode, the fuzzy rule table is shown in Table 4. Their membership functions are shown in Fig. 8.
2.2.4 Velocity control

Ultrasonic Sensors

Dynamic Switching

Goal-Seeking Mode

Obstacle Avoidance Mode

Wall Following Mode

Steering Coefficient Average Speed Y'avg

Right Wheel Velocity: T 'L

Wheel Velocity: V

Fig. 4 Operation of the navigation agents

The velocity control is important to the navigation modules, and it promotes a comfort wheelchair driving. It is inferred in terms of the steering coefficient (t , an output from previous navigation modes) and the minimal distance (dist) of D_ target and D_obstacle. Where D_target denotes the distance between the wheelchair and the target; and D_obstacle denotes the distance between the wheelchair and the front obstacle. It aims to increase the safety and comfort of steering. In this manner, increasing steering angle will result in the decreasing in speed of the wheelchair. The fuzzy rule table is shown in Table 5. Where Vavg is the average linear speed of two rear wheels. Their membership functions are shown in Fig. 9.

3. RESULTS
In this work, an agent-based control algorithm of a robotic wheelchair is discussed. To verify the performance of the proposed control algorithm, the mechanical geometry specification of a simulated robotic wheelchair is designed as shown in Fig. 10. The diameter of the rear wheel is 410 mm; the distance between two rear wheels is 750 mm; and the length of the wheelchair is 780 mm, as indicated in Fig. 3. The kinemttics is also derived as shown in Fig. 11. Two wheelchair motion cases are discussed as in Fig. If. The first case (a) indicates the robotic wheelchair rotating two wheels in the same direction; and case (b) indicates the robotic wheelchair rotating two wheels in the opposite directions. Where Bo indicates original vehicle angle; ON indicates new

Centre of two rear wheels If(a-I)> 180,then 0 =a-Il-360 If (a - Il) <- 180, then 0 = a - I -360 Otherwise, 0 = a - I Where 0<a, P <360 ; and -180 < 0 <180'

Fig. 5 Goal-seeking input parameter definition

-15-

227

Vol. 15 No. 6 December 2003

Table 1 Dynamic fuzzy fusion method switch table ul U2 u3 x X X X X X X 0 0 0

-

u4

u5 0 0 X 0 X 0 0 0 -

u6 x 0

u7 x

Use Mode Goal- Seeking Goal-Seeking Goal-Seeking Goal-Seeking Wall-Following Wall-Following Wall-Following Wall-Following Wall-Following Wall-Following
Obstacle Avoidance

x x
X

X
X X X

x
X X

X
X X

X
0

x
X X

X
0

X
X X X X

X
0 0 0 0
-

X
0 0 0 0

x
X X X
O

X
X X X
-

X
-

O O O

0 0 0 0

0 0 0 0 -

Obstacle Avoidance Obstacle Avoidance Obstacle Avoidance Obstacle Avoidance Obstacle Avoidance Obstacle Avoidance

Table 2 . Goal-seeking mode fuzzy rule table 0 It NB NB NM NM NS NS ZE ZE PS PS PM PM PB PB

vehicle angle; 0 indicates the original vehicle center position; N indicates the original vehicle center position; o0 indicates the rotate angel; D indicates the distance between two wheels; T indicates the instance center; R indicates the distance from right wheel to point "T"; dL indicates the displacement of the left wheel; and dR indicates the displacement of the left wheel. Practically, this paper uses the Microsoft Visual C++ [7] and SQL server to develop the application softwares. The individual program acts as the distributed agent module. The agent modules are intercommunicated in terms of the Windows Socket. The following cases are the simulation results of various navigation environment to validate the proposed distributed agent-based control modules. Initially, the validation of uisng only one navigation agent module alone is discussed based on the goalseeking, obstacle avoidance, and wall-following, respectively. The simulation result of using the "obstacle

avoidance" agent only is shown in Fig. 12. The result shows that the wheelchair moves freely without the knowledge of the environment map and all obstacles are avoided. Fig. 13 shows the the simulation result of using the "wall-following" agent only. In this figure, the wheelchair moves alone the specified wall without the knowledge of the environment map. The result shows that the wheelchair keeps a constant distance awary from the wall. Consequently, the "wallfollowing" agent is validated. In addition, the simulation result of using the "goal-seeking" agent only is also discussed, as shown in Fig. 14. In this figure, the wheelchair moves from the center to eight different targets with initially in the north direction. The result shows that the wheelchair reaches the targets precisely. In addition, since the circles drawn in this figure indicate the wheelchair positions recorded in a specified sampling time, the velocity of the wheelchair can be investigated from the density of the circles. If the circles are drawn in high density, then they indicate the robotic wheelchair

-16-

BIOMEDICAL ENGINEERINGAPPLICATIONS, BASIS & COMMUNICATIONS

228 moving slowly. Consequently, the wheelchair moves slowly when it turns in a large angle or it reaches its destination. Such a control behavior ensures a comfort motion drive. In addition, the cooperation between the velocity control agent and the goal-seeking agent is also validated. Based on previous discussion, there major navigation agent modules were validated. Since these navigation agent modules are developed based on the simplified environment, they can not be applied to the real complex and dynamic environment. In this work, the fuzzy fusion method is proposed to resolve such problems based on the dynamic switching table as shown in Table 1. The fuzzy fusion method can be desired in terms of dynamic switching of the the validated navigation agent modules. Fig. 15 shows the results of the fusion of the "goal-seeking" and "obstacle avoidance" agent modules. In this figure, the robotic wheelchair controller initially operates using the "goal-seeking" agent module. Once the ultrasonic sensor detects the obstacles, the controller switches to the "obstacle avoidance" agent module. When the obstacles are not in front of the wheelchair, the controller automaically changes to the "goal-seeking" agent module again. In addition, the operatoion of the velocity control agent module is also validated. Similarly, the simulation result of the fusion of the "wall-following" and "obstacle avoidance" agent modules is shown in Fig. 16. In this figure, the robotic wheelchair controller initially operates using the "wallfollowing" agent module based on the inference table as indicated in Table 1. Once the ultrasonic sensor detects the obstacles, the controller switches to the "obstacle avoidance" agent module. When the obstacles are not in front of the wheelchair, the controller automaically changes to the "wall-

XW

(a)

Fig. 6 Goal-seeking mode membership functions

0.5

0 0

a
NB

60

80

100

120

140

160

80

200

Input membership function (Dist)

NS PS PMPB

0.5

X><XX><X-80
-100
-60 -40 -20 0 20 40 60

so

100

Output membership function ()

Fig. 7 Obstacle avoidance mode membership functions

Membership function of Input parameter,

10

20

30

40

50

60

70

80

10(
0.5

Input membership function (Dist)

NM NS PS PM
P

Membership function of Input parameter, dist

0

00

150

0
-100 -80 60 -40 -20 0 20 40 60 80
1

Membership function of Output parameter, V....

10 50 60

00

Output membership function ()

Fig. 8 Wall-following mode membership functions

Fig. 9 Velocity control membership functions

-17-

229

Vol. 15 No. 6 December 2003

Table 3 . Obstacle avoidance mode fuzzy rule table

1) Condition 1: if Dsl is Small (S)

Ds3 P S M Ds2 B L PB PB PB PB S PB PB M PB PB B PB PB PB PB L PB PB PB PB

(2) Condition 2: if Dsl is Medium (M)

Ds3 S S Ds2 M B L (3) Condition 3: if Dsl is Big (B) Ds3 S S M Ds2 B L NB NB NM NM PM PM PM PS NB NB M NB NB B PB PB L PB PB PB NB NB NB M PB PB PB PB B PB PB PM PM L PB PB PM PM

(4) Condition 4: if Ds 1 is Large (L)

Ds3 S S Ds2 M B L NB NB NB NB M NB NB NM NM B NB NB NM NS L PB PB PM ZE

following" agent module again. In addition to the fusion of any two agent modules, three navigation modules can also be automatically fused. Fig. 17 is a specially designed environment map, and it can be used to validate the fusion of three navigation modules. Initially, the controller uses the goal-seeking agent, and then

switches to the obstacle avoidance agent. Next, the controller switches to the wall-following agent module. Then, the controller switches to the goal-seeking agent module again. Similarly, the other switches can be indicated in Fig. 17. Finally, the wheelchair reaches its destination. It is noted again, the validation of velocity control agent module can also be investigated in terms

-18-

BIOMEDICAL ENGINEERINGAPPLICATIONS, BASIS & COMMUNICATIONS

230

Table 4 . Wall-following mode fuzzy rule table

(1) Condition 1: follow right wall

Ds4 S
S M Ds5

M PS ZE NS NM

B PM PS PS ZE

L PM PM PM PM

NS NS NM NM

B L (2) Condition 2: follow right wall

Ds7

S S Ds6 M
B L Table 5. Velocity control fuzzy rule table

M NS ZE PS PM

B NM NS NS ZE

L NM NM NM NM

PS PS PM PM

disc
Vavg.

S NB NM NS ZE PS
PM

M ZE ZE PS PS PS
ZE

B PS PS PM PM PM
PS

L PM PM PB PB PB
PM

ZE ZE ZE ZE ZE
ZE

PB

ZE

ZE

PS

PM

of the circle density.

The previous examples validated the stationary environments. Fig. 18 validates the navigation of the dynamic environment. The rectangle blocks indicate the dynamic obstacles. In this figure, the wheelchair reaches its destination in terms of dynamically switching the goal-seeking, obstacle avoidance, and wall-following agent modules, and it can also reach its destination precisely.

and without using the path planning agent module. Apparently, the wheelchair path of using the A* path agent module is better. The path agent module generates several sub-goals based on the knowledge of the environment map. The wheelchair controller executes the sub-paths that is generated from the subgoal sequences. It is noted that any static or dynamic obstacles appeared on the sub-path will exeute the agent module fusion method to aviod collision.

On the other hand, the A* path planing agent module is validated. Fig. 19 shows the results of uisng

Finally, the coordination of multi-wheelchair environment is simulated and validated. Fig. 20 shows

-19-

231

Vol. 15 No. 6 December 2003

Fig. 10 Simulated robotic wheelchair 3D diagram Fig. 12 Example of using obstacle avoidance agent only

Ae-d^-a.
D R_ d"

AO
(a)

OT-D+R 2 T (X,, Y,) - T(x, +OT cos(O0 - M), yo +OT sm( Oo - 2 )) N(x,, yw) _ N(x, +OT cos(B, + Z -AO ), y, +OTT sin(Oo + 2 - AO)) O, -Oo -AO

Fig. 13 Example of using wall-following agent only

R

e8_d^+d D OT-(2 1 d,, d+R dd)D T(xr., yr) - T (x, +OT cos(O0 - 2 ), Y, +OT sin(o - 2)
)

priority wheelchair reaching its destination as soon as possible.

4. DISCUSSION
In this paper, the agent-based robotic wheelchair control system is developed. In order to validate the proposed agent-based robotic wheelchair controller, a simulated robotic wheelchair and several static and dynamic environments are constructed. These environments are analogous to the clinical environments. This work aims to promote a safe and comfortable wheelchair operation environment. Therefore, the significances of this work can be discussed as follows: 1. The agent-based wheelchair control architecture

N(xr, y,) - N(xr +OT cos(O, + '- AO), Yr +OT sin(Oo + 'g- AO))

eH -eo -AO (b)

Fig. 11 Kinematics of the robotic wheelchair the collision avoidance with priority control of two robotic wheelchairs. In this case, since the wheelchair A has higher priority (assigned by system coordination agent) than wheelchair B, wheelchair B parries away from the path of wheelchair A when their path intersect. Such a control mechanism ensures the higher

-20-

BIOMEDICAL ENGINEERINGAPPLICATIONS, BASIS & COMMUNICATIONS

232

Fig. 14 Example of using goal-seeking agent only

Fig. 17 Module fusion example of three navigation agent modules

san n

Fig. 15 Module fusion example of goal -seeking and obstacle avoidance

Fig. 18 Module fusion example of the dynamic obstacles is constructed based on three-layer hierarchy. Each wheelchair control module is developed as the agent, and the individual agent module can be implemented and validated independently. Such system development architecture increases the integrability and maintability. 2. The proposed wall-following, goal-seeking and obstacle avoidance navigation agent modules are developed in terms of the practical considerations. It aims to assist the disabled patients or elders to move and to reach the desired location targets by avoiding collisions. 3. In order to operate the wheelchair in complexly clinical environments, the navigation agents are fused together to propose dynamically switching mechanism. The fuzzy fusion agent module determines a suitable navigation agent

Fig. 16 Module fusion example of wall-following and obstacle avoidance

-21-

233

Vol. 15 No. 6 December 2003 integration is easy and straightforward. Therefore, the agent technology is beneficial to solve the robotic wheelchair control problems. 7. Finally, 9 stationary and dynamic environments are discussed and validated based on the clinical considerations successfully. In the future, they can be further modified for the clinical validation.

5. CONCLUSION
In this paper, a robotic wheelchair controller is developed based on the multi-agent control architecture and fuzzy algorithms. All the fuzzy agent modules are validated in this paper. In addition, the fuzzy fusion method is proposed to resolve the complexly clinical environment navigation problems. These navigation agent modules can be cooperated with the velocity control agent. Therefore, the proposed robotic wheelchair controller assists the disabled patients or elders to reach the desired location targets with comfort and collision free. On the other hand, the path planning and sub-goal agent modules are used to reduce the wheelchair path. Finally, based on various validations, the proposed robotic wheelchair controller achieves the desired objectives. In the future, the wheelchair and the proposed controller can be implemented physically to valid the clinical usability and performance.

Fig. 19 Module fusion example of using A* path planning agent module

ACKNOWLEDGEMENT
Fig. 20 Collision avoidance with priority control of two robotic wheelchairs module to operate in terms of reasoning the ultrasonic sensor signals. In this manner, the robotic wheelchair can reach its targets with higher performance. 4. In this work, the velocity control agent is cooperated with any of the navigation agent to perform comfortable wheelchair driving.
5. The A* path planning agent can use the preknown environment map to generate the optimal paths and the corresponding sub-goal sequences. The sub-goal sequences can be further transformed to the sub-path sequences, and the navigation agents execute the sub-path sequentially. Therefore, the path planning and sub-goal agents reduce the distance of wheelchair path.

This work was partially supported by the National Science Council, Taiwan, R.O.C., under Grants NSC89-2212-E-182-013. In addition, the authors appreciate the technique supports from RealScene Technology Corp. (DC motor drive), Intelligent Control Technology Co., Ltd. (DSP system), and UPTECH Technology Co., Ltd. (wireless modem).

REFERENCES
1. H. Asai, "Motion Planning of Obstacle Avoidance Based on Guide Style for Autonomous Mobile Robot", Proceedings of the Seventh International Conference on Virtual Systems and Multimedia (VSMM'01), Vol. 8, pp. 694 -700, 2001. 2. G. Bourhis, O.Horn, O.Habert, A.Pruski, "An Autonomous Vehicle for People with Motor Disabilities" IEEE Robotics & Automation Magazine, Vol. 8, pp. 20 -28, 2001.
3. G. Bourhis, O. Horn, O. Habert, A. Pruski, "The Autonomous Mobile Robot SENARIO: A sensor

6. This work develops several path planning, navigation, and velocity control algorithms. These algorithms must be integrated to propose an overall robotic wheelchair controller. By using the agent based control architecture, the

-22-

BIOMEDICAL ENGINEERINGAPPLICATIONS, BASIS & COMMUNICATIONS Aided Intelligent Navigation System for Powered Wheelchairs" IEEE Robotics & Automation Magazine, Vol. 4, pp. 60-70, 1997. 4. H. Fiorino and C. Tessier, "Agent Cooperation: a Petri Net Based Model," Proceedings of International Conference on Multi Agent Systems, pp. 425-426, 1998.

234 11. H. Maaref and C. Barret, "Sensor-Based Fuzzy Navigation of an Autonomous Mobile Robot in an Indoor Environment", Control Engineering Practice, vol. 8, pp. 757-768, 2000. 12. M. Mazo, "An Integral System for Assisted Mobility" IEEE Robotics & Automation Magazine, Vol. 8, pp. 46 -56, 2001. 13. M. Mucientes, R. Iglesias, C.V. Regueiro, A. Bugarin, P. Carinena, and S. Barro, "Fuzzy Temporal Rules for Mobile Robot Guidance in Dynamic Environments", IEEE Transactions on Systems, Man, and Cybernetics-Part C, Vol. 31, pp. 391-398, 2001. 14. N.J. Nilsson, "Artificial Intelligence: A New Synthesis," Morgan Kaufmann Publishers, 1998. 15. B.S. Ryu and H.S. Yang, "Integration of Reactive Behaviors and Enhanced Topological Map for Robust Mobile Robot Navigation", IEEE Transactions on Systems, Man, and CyberneticsPart A, Vol. 29, pp. 474-485, 1999.

5. H. Hiraishi, H. Ohwada and F. Mizoguchi, "WebBased Communication and Control for Multiagent Robots," Proceedings of 1998 IEEE/RSJ International Conference on Intelligent Robots and Systems, Vol. 1, pp. 120-125, 1998. 6. T. Holvoet and P. Verbaeten, " Using Agent for Simulation and Implementation Petri Net, " Proceeding of 11th Workshop on Parallel and Distributed Simulation, pp. 134-137, 1997.
7. D.J. Kruglinski, Inside Visual C++ - The Standard Reference for Programming with Microsoft Visual C++ Version 4, Washington:Microsoft, 1996. 8. A. Lankenau, T. Rofer, "A Versatile and Safe Mobility Assistant" IEEE Robotics & Automation Magazine, Vol. 8, pp. 29 -37, 2001. 9. R.C. Luo and T.M. Chen, "Development of a Multibehavior-Based Mobile Robot for Remote Supervisory Control Through the Internet", IEEE/ASME Transactions on Mechatronics, Vol. 5, pp. 376-385, 2000. 10. R.C. Luo, T.M. Chen, M.H. Lin, "Automatic Guided Intelligent Wheelchair System Using Hierarchical Grey-Fuzzy Motion Decision-Making Algorithms", Proceedings of the 1999 IEEE/RSJ International Conference on Intelligent Robots and Systems, pp. 900-905, 1999.

16. A.V. Timofeev, F.A. Kolushev, A.A. Bogdanov, "Hybrid Algorithms of Multi-Agent Control of Mobile Robots", IJCNN International Joint Conference on Neural Networks, vol. 6, pp. 4115 4118, 1999.
17. S.G. Tzafestas, "Research on Autonomous Robotic Wheelchairs in Europe" IEEE Robotics & Automation Magazine, Vol. 8, pp. 4 -6, 2001.

18. M. Wooldridge and M.R. Jenkins, "Intelligent Agents: Theory and Practice", Knowledge Engineering Review, pp. 115-152, 1995. 19. Microsoft ODBC 2.0 Programmer's Reference and SDK Guide, Microsoft, 1994. 20. http://mdnews.itri.org.tw/

-23-