MULTI-LEVEL CONTROL OF AN INTELLIGENT

WHEELCHAIR IN A HOSPITAL ENVIRONMENT USING A

CYBER-MOUSE SIMULATION SYSTEM

Rodrigo A. M. Braga, Marcelo Petry, Eugenio Oliveira and Luis Paulo Reis

Artificial Intelligence and Computer Science Lab-LIACC, Faculty of Engineering of University of Porto

Rua Dr. Roberto Frias, s/n 4200-465, Porto, Portugal

Keywords: Intelligent Wheelchair, Intelligent Robotics, Strips Planning, Path Planning, A* Algorithm.

Abstract: The development of intelligent wheelchairs is a very good solution to assist severely handicapped people

who are unable to operate classical electrical wheelchair by themselves in their daily activities. This paper

describes the integration of a robotic simulator with our intelligent wheelchair shared control and planning

modules. An adapted version of the free Cyber-Mouse robotic simulator was used to simulate the movement

of the intelligent wheelchair in a hospital environment. Adaptations of the subsumption architecture, Strips

Planning and A* Algorithms were employed and integrated to allow wheelchair intelligent behavior. The

experimental results have demonstrated the success of the integration of these algorithms in our simulator

allowing very safe motion of the intelligent wheelchair in the simulated hospital environment. Also, the

adapted Cyber-Mouse simulator proved its capability and robustness in simulating the hospital environment

and wheelchair physic characteristics.

1 INTRODUCTION

Wheelchairs are important locomotion devices for

handicapped and senior people. With the increase in

the number of senior citizens and the increment of

people bearing physical deficiencies in the social

activities, there is a growing demand for safety and

comfortable Intelligent Wheelchairs (IW) to

practical uses. The main functions of IWs are (Faria,

2007a)(Faria, 2007b)(Jia, 2005):

•Interaction with the user, including hand based

control, such as, joystick, keyboard, mouse, touch

screen; voice based control, such as audio; vision

based control, such as camera; and other sensor

based control, such as pressure sensors.

•Afford Services, for instance autonomous

navigation (with safety, flexibility and robust

obstacle avoidance), communication with other

devices (like automatic doors).

This paper discusses the application of a Cyber-

Mouse simulator, developed at the Univ.Aveiro in

Portugal (Lau, 2002) in the study, development and

test of shared control and high-level planning

algorithms applied in an IW operating in a hospital

environment.

A shared control algorithm was tested, allowing

IW automatically avoids danger situations. Also,

typical algorithms used in most intelligent robotics

applications were applied in the control of the IW

and simulated in the hospital scenery. Blended with

the control, a motion planner was developed capable

of generating the behavior/path commands

according to an a-priori map of the world. This

motion planner is capable of instructing the low-

level motion controller module to achieve the high-

level commands desired by the user (Luo, 1999).

Cyber-Mouse is a competition among virtual

robots, which takes place in a simulated

environment running in a computer network. The

simulation system creates a virtual arena with a

starting grid, a target area, signalled by a beacon,

and populated with obstacles. It also creates the

virtual bodies of the robots. The simulator estimates

sensor measures which are sent to the agents.

Reversely, receives and apply actuating orders

coming from agents (Lau, 2002).

The paper is subdivided in 5 different sections:

section 2 presents some brief definitions of the

control algorithms applied in this research; section 3

contains a description of the system developed;

section 4 provides experimental tests and result

discussion and section 5 presents the final

conclusions and points out some future research

topics.

179

A. M. Braga R., Petry M., Oliveira E. and Paulo Reis L. (2008).

MULTI-LEVEL CONTROL OF AN INTELLIGENT WHEELCHAIR IN A HOSPITAL ENVIRONMENT USING A CYBER-MOUSE SIMULATION SYSTEM.

In Proceedings of the Fifth International Conference on Informatics in Control, Automation and Robotics - RA, pages 179-182

DOI: 10.5220/0001498801790182

 SciTePress

2 INTELLIGENT WHEELCHAIR

CONTROL ALGORITHMS

To test the simulator, the algorithms used to

implement the control system and algorithms used

for the high-level planning system were:

- Adapted Subsumption Architecture. (Brooks,

1991) (Russell, 2002)(Ferber, 1991) for basic

moving;

- Knowledge Representation using Grids.

Spatial knowledge for robotic movement,

navigation and planning (Borenstein,

1991)(Thrun, 1996).

- A* Search Algorithm for Trajectory

Generation. used to find a path from a given

initial node to a given goal node (Shapiro, 2000)

(Hart, 1968) (Barr, 1986).

- Adapted Strips Planning Algorithm. Planning

is the task of searching for an action sequence to

achieve a given goal (Russell, 2002)(Weld,

1999)(Fikes, 1971) (Bonet, 2001).

3 SYSTEM DESCRIPTION

The hospital environment was created modifying the

Cyber-Mouse labyrinth definition XML file that

contains positions of a set of walls of arbitrary shape

and beacons that emit signals. Figure 1 shows our

hospital modified floor plan.

Figure 1: Hospital environment represented using the

Cyber-Mouse simulator.

The hospital representation is composed of five

rooms: bathroom, kitchen, hall and garden and a set

of walls of different shapes. IWs are represented by

modified robotic agents. UDP Sockets are used for

communication between the robotics agents and the

simulator. The communication is based in

dispatching XML messages of five types: request

action order.

Figure 2 presents the developed agent control

software interface. A communication configuration

panel can be seen in left side of the interface, the

sensor values are depicted in the middle of the

interface. In the right side, there are three function

modules. The first module has the following

functions: semi-automatic control, revolve in a point

direction, move to a point, move through planning

path.

Figure 2: Intelligent Wheelchair software interface.

The semi-automatic control, also denoted shared

control, was constructed with fusion between user

command and subsumption architecture. This way it

achieves safe movement allowing automatic obstacle

avoidance. Figure 3 presents the implemented

architecture.

Figure 3: Subsumption architecture used in shared control.

Figure 4 shows the path generation interface

module. The path generation was implemented using

the A* algorithm. The state world information is

loaded into the system using a XML file. A grid of

56x28 cells representing the world- state, for this

development, was considered as the world state

environment. Also, object extension techniques were

used, in such manner that a robot may be

represented as a point in the system without the risk

of collision with the walls.

The path generation module searches for a path,

starting in the actual point and finishing in the user

selected point. The path generated can be seen in the

interface provided.

The last module (see Figure 5) is an automatic

planning module developed based in Strips Planning

Algorithm with inverse chaining. Initially, we

developed five general high-level objectives: “go-

bathroom”, “go-garden”, “go-room”, “go-

kitchen”, “go-consulting_room”. These general

objectives generate different final objectives,

depending on the patient, wheelchair and location

set for the actions. For example, the action patient 1

with wheelchair 2 go-room for Room 1 results in the

objective: patient 1 in Room 1, patient 1 without

wheelchair, wheelchair 2 free and situated in hall.

In this way, it is allowed to set many different final

objectives and since the plan depends on the initial

ICINCO 2008 - International Conference on Informatics in Control, Automation and Robotics

180

state, it will have different sequences and number of

actions for arriving the final objective.

Figure 4: Path generation module interface.

After plan generation, we can execute it in the

simulated environment. Now, the system uses the

path generation module for the generation the paths

for movement actions and automatic move control

module to execute the movement through the path

generated with collision avoidance. This module

executes step by step the actions of the plan using

the other modules whenever it is necessary.

Figure 5: Planning module interface.

4 EXPERIMENTS AND RESULTS

This section presents a simple operation example of

our robotic agent and results gathered in this

experiment. After initiating the simulator, the

configuration, map and grid file are loaded in the

simulator, through a XML file.

As focus is primarily testing the simulation

system and the implementation of an ordinary multi

level control, sensor and motor noise were disable.

Another simplification is the robot data position,

which is provided by the simulator GPS system that

represents a robust odometry system.

The simulator configuration is presented in

Table 1. The simulator allows inserting noise in the

robots sensors and actuators. However in this stage

of development, the sensor and actuators noise was

not used.

The first test was the use of the A* algorithm for

searching the path from point (x=9,5 and y=13,2) to

point (x=14 and y=7). The plan devised is presented

in Table 3. The plan is presented in inverse order,

i.e., the first point is in the table bottom and last

point is in the table top.

Table 1: XML file to configure simulator.

<Parameters SimTime="18000" CycleTime="80" CompassNoise="0"

BeaconNoise="0" ObstacleNoise="0" MotorsNoise="0"

RunningTimeout="1350" GPS="On" ShowActions="False" Nbeacons="5"

Lab="C:/CiberToolsWindows-1.2.0/Labs/My2Lab.xml"

Grid="C:/CiberToolsWindows1.2.0/Labs/MyGrid.xml" />

The resulting movements are shown in Figure 6.

The black letters represent before of move, the blue

line represent plan path and the red line represent the

path travelled. The number ‘0’ represents free space,

‘1’ represents the expanded wall. In this test, the

move was free collision and with minimum

trajectory error.

Figure 6: Movement result.

Next step was to generate a comprehensive plan

using the implemented path generator. The goal in

this case was to pick up patient 1 in bedroom 1;

carry him to bedroom 5, to make company to patient

5; and finally go back to the lobby. In this example

we had as objective: “Go to the bedroom”, as Place:

5 and Patient: 1. This parameters result as final

objective:

On(P1,R5) ^ WithoutC(P1) ^ On(C2,H). The

world state before the action, the resulting plan and

the following state are represented in Figure 7.

Figure 7: World state s, plan and world state s'.

The grid containing the final state and the

travelled path, based on the plan mentioned before,

can be observed in Figure 8. This grid contains the

four basic movements that the IW needed to perform

to achieve the final objective: from the initial point

until the lobby, from lobby to bedroom 1 (to pick up

patient 1), then to the bedroom 5 and after leave

patient 1 going back (empty) to the lobby. The time

to conclude movement in automatic control was 982

second and for manual control was 870 seconds.

Figure 8: Final movement result.

MULTI-LEVEL CONTROL OF AN INTELLIGENT WHEELCHAIR IN A HOSPITAL ENVIRONMENT USING A

CYBER-MOUSE SIMULATION SYSTEM

181

5 CONCLUSIONS

This paper presented the implementation of a robotic

Intelligent Wheelchair, simulating its behavior in an

adapted Cyber-Mouse simulator. This agent resulted

of a modular system composed of three modules.

The first one is a shared control that merges user

commands with the information received by the IW

sensors. This advantage prevents the collision with

objects in its way becoming movement system more

reliable and safe. The algorithms developed use a

subsumption architecture: once there are objects

close to the wheelchair, the user commands are

disabled and deflect commands are sent to the

motors, ensuring a safe motion.

A module was develop using the A* algorithm as

path generator to calculate the shortest path from the

robot’s actual position to the objective point. The

third module was implemented to plan some usual

tasks in a hospital environment, using Strips

planning to solve these interactions. In most of the

cases Strips proved to be efficient, delivering

complete plans, with exceptions presented under

Sussman anomaly effects, where one operation

cancels other in the plan.

It was also demonstrated that Cyber-Mouse is a

friendly tool to test control algorithms, IW

navigation and its interaction with hospital

environment.

Future research directions include the

improvement to non-linear planning and upgrade

from A* to D* algorithm, once it’s preferable in

such a dynamic environment. To be fully intelligent,

it’s not enough the wheelchair to plan its own path

or share its control. It is also necessary to

communicate with other intelligent wheelchairs and

devices like doors activation systems, elevators and

lights. Due to this, it is intended to perform an in

depth study in proper methodologies to implement

these capacities in the wheelchair, and this way,

implement some functions as cooperative behavior

among a group of IW and collaboration among the

user and the system. In Cyber-Mouse it is necessary

to increase its present simulation capacities, from its

actual three IW, enabling hybrid systems test, where

real and virtual IW interact with each other. These

interactions make possible high complexity tests

with a substantial number of devices and

wheelchairs, representing a reduction in the project

costs, once there wouldn’t be necessary a large

number of real IW. Still, in Cyber-Mouse, it is

necessary to implement noise treatment in the

motors and sensors to have actions in the simulated

wheelchair closer to those of real wheelchairs.

REFERENCES

Barr, A. Cohen, P. Feigenbaum, E., 1986. Handbook of

Artificial Intelligence. Addison Wesley. Vol. 1.

Bonet, B. and Geffner, H., 2001. Heuristic search planner

2.0. The AI Magazine, 22(1), pp.77–80.

Borenstein, J., Koren, Y., 1991. The Vector Field

Histogram - Fast Obstacle Avoidance for Mobile

Robots, IEEE Journal of Robotic and Automation, v.

7, n. 3, pp. 278-288.

Brooks, R. A., 1991. Intelligence Without Reason. Proc.

12th Int. Joint Conference on Artificial Intelligence -

IJCAI-91, Sydney, Australia. pp.569-595.

Faria, P. M., Braga, R A. M., Valgôde, E., Reis L. P.,

2007a. Platform to Drive an IW using Facial

Expressions. Proc. 9th Int. Conf. on Enterprise

Information Systems, HCI, ICEIS’2007. pp.164-169.

Funchal, Portugal. June 12–16.

Faria, P. M., Braga, R. A. M., Valgôde, E., Reis, L. P.,

2007b. Interface Framework to Drive an Intelligent

Wheelchair Using Facial Expressions. IEEE Int.

Symp. on Industrial Electronics, pp. 1791-1796, June

4-7.

Ferber, J.; Gasser, L., 1991. Intelligence artificielle

distribuée. In: Int. WorkShop on Expert Systems &

Their Applications, 10., Avignon. Cours n. 9. France.

Fikes, R., Nilsson, N. J., 1971. STRIPS: A New Approach

to the Application of Theorem Proving to Problem

Solving. IJCAI 1971. pp.608-620.

Hart, P. E., Nilsson, N. J. and Bertram R., 1968, A formal

basis for the heuristic determination of minimum cost

paths. IEEE Transactions on Systems Science and

Cybernetics, 4(2):pp.100–107.

Jia, P. and Hu, H., 2005. Head Gesture based Control of

an Intelligent Wheelchair. CACSUK-11

Ann. Conf.

Chinese Aut. Comp. Society in the UK, Sheffield, UK.

Lau, N., Pereira, A., Melo, A., Neves, A. and Figueiredo,

J., 2002. Ciber-Rato: Um Ambiente de Simulação de

Robots Móveis e Autónomos. Rev. DETUA, Vol. 3,

Nr. 7, pp.647-650, Sept.

Luo, R. C., Chen, T. M. and Lin, M. H., 1999. Automatic

Guided IW System Using Hierarchical Grey-Fuzzy

Motion Decision-Making Algorithms, Proc. IEEVRSJ

1999 Int. Conf. on Intelligent Robots and Systems.

Russell, S. and Norvig, P., 2002. Artificial Intelligence: A

Modern Approach. Prentice Hall, E.Cliffs, NJ, 2

ed.

Shapiro S.C., 2000. Encyclopedia of Artificial

Intelligence. Wiley-Interscience. May.

Thrun, S. and Bücken, A., 1996. Integrating Grid-Based

and Topological Maps for Mobile Robot Navigation,

In: Proc. 13th National Conference on Artificial

Intelligence AAAI, Portland, Oregon, pp.944-950.

Weld, D. S., 1999. Recent advances in AI planning. AI

Magazine, pp.93–123

ICINCO 2008 - International Conference on Informatics in Control, Automation and Robotics

182