ACTION CONTROL METHOD FOR POWERED WHEELCHAIRS

CONSIDERING CONTROL INPUT

AND ENVIRONMENTAL INFORMATION

Yu Ishihara

Graduate School of Science and Technology, Keio University, 3-14-1, Hiyoshi, Kohoku-ku, Yokohama, Japan

Masaki Takahashi

Department of System Design Engineering, Keio University, 3-14-1, Hiyoshi, Kohoku-ku, Yokohama, Japan

Keywords: Powered Wheelchair, Human Robot Interaction, Corridor Detection.

Abstract: Powered wheelchair is a mobility commonly used for people with disabilities. Recently, to improve the

safety and efficiency of powered wheelchairs, various systems with control assistance are proposed.

However most of the systems suppose joystick as input device and doesn’t consider about people with

difficulties using it. Not all devices alternative to joystick have enough operability compared to joystick. In

this paper, considering corridor passing as verification environment, an action control method to drive

through the corridor safely and efficiently for wheelchairs controlled by devices with low operability is

presented. To achieve safe and efficient driving, proposed method considers the time series of passenger’s

input commands and combines the environmental information to select effective direction and speed for the

wheelchair instead of passenger. Moreover for environment recognition, corridor detection algorithm is also

proposed. To verify the effectiveness of proposed method, simulations and experiments were carried out.

1 INTRODUCTION

Recently, the number of wheelchair users is

increasing at an average annual rate of 5.9 % a year

and who have benefit from smart wheelchair,

wheelchair with assist facility, is estimated as 61 to

91 % of all wheelchair users (Simpson, 2008).

Hence assist facility is becoming one of the

important features for users with disabilities that find

difficulties in operating existing powered

wheelchairs. In order to facilitate the operation,

several wheelchairs were developed with obstacle

avoidance methods and/or passenger’s intention

estimation so far and their effectiveness were

verified (Levine, 1999); (Iturrate, 2009). Though

most of the previous methods were focused on the

wheelchair controlled by joystick and didn’t

consider people with difficulties using joystick: e.g.

people with handicaps on upper limbs. Therefore

wheelchairs enabled to control by devices alternative

to joystick such as voice command (Simpson, 2002),

face recognition (Saitoh, 2007) and brain machine

interface (BMI) (Vanacker, 2007); (Iturrate, 2009)

are recently proposed. However most of these

devices don’t have enough operability, low

operating frequency, few input direction and

uncertainty of information, compared to joysticks

and needs further assist to establish safe and

efficient control of wheelchair as shown in Figure 1.

Previously, two approaches are proposed to

control the wheelchair with low operability devices

(Vanaker, 2007); (Iturrate, 2009). One is automatic

navigation and the other is implementation of

existing approaches. Automatic navigation allows

the passenger not to control directly the wheelchair.

Though this approach limits the wheelchairs travel

range of passenger as the navigation needs the map

of environment preliminary. And implementation of

existing methods enable to control the wheelchair

safely but its efficiency is not same as that of

joystick. For example, it is reported that the

wheelchair controlled by BMI has difficulty in

driving the wheelchair as shown in Figure 2

(Vanaker, 2007).

Ishihara Y. and Takahashi M..

ACTION CONTROL METHOD FOR POWERED WHEELCHAIRS CONSIDERING CONTROL INPUT AND ENVIRONMENTAL INFORMATION.

DOI: 10.5220/0003534500280036

In Proceedings of the 8th International Conference on Informatics in Control, Automation and Robotics (ICINCO-2011), pages 28-36

ISBN: 978-989-8425-75-1

 2011 SCITEPRESS (Science and Technology Publications, Lda.)

Figure 1: The relationship between the operability of

device and necessity of assist facility.

Figure 2: Inefficient driving of wheelchair by brain

machine interface due to the low operability.

Therefore in order to control the wheelchair with

devices alternative to joystick safely and efficiently,

the assist facility has to take care of the features of

the device and needs further assist of passenger’s

operation during the travel. More specifically, to

achieve efficient and safe control of wheelchair

without environmental constraint, reactive action

control that comprises the devices’ limitations is

necessary.

In this paper, regarding corridor passing as a

general task for wheelchair users, a new action

control method to drive through the corridor safely

and efficiently by devices with low operability is

presented. For simplicity, we suppose that the input

information doesn’t have uncertainty. Particularly,

we consider for the next two points:

(a) Deceleration before entering to the cross roads

(b) Smooth and efficient turns to change direction

(a) is to improve the safety of passenger from the

undetectable moving objects shield by the walls like

in Figure 3 and for the difficulty of keeping

particular speed using low operability devices. (b) is

to achieve efficient driving and to prevent from

stopping at all times changing direction.

To achieve the points mentioned above, we propose

a corridor detection algorithm and an action control

method that simultaneously considers control input

and environmental information. The wheelchair first

considers the time series of passenger’s inputs to

deal with the low frequency and decides roughly the

moving direction. Next, by combining the

environmental information with decided direction,

proposed method analyze the passenger’s intention

and selects effective direction and speed for the

wheelchair instead of passenger. Finally combining

the selected direction and speed with obstacle

avoidance methods, proposed method realizes safe

and efficient driving of wheelchair. To verify the

effectiveness of the proposed method, several

Figure 3: Example situation that deceleration is needed.

Figure 4: Flowchart of proposed algorithm.

simulations and experiments were carried out.

2 CONTROL SYSTEM

OF WHEELCHAIR

The control system of wheelchair consists of three

steps, corridor detection, evaluation of input

commands and action control. The flowchart of our

proposed method is shown in Figure 4. By

considering the input commands of passenger and

environmental information such as corridors and

ACTION CONTROL METHOD FOR POWERED WHEELCHAIRS CONSIDERING CONTROL INPUT AND

ENVIRONMENTAL INFORMATION

obstacles to decide the action, the system achieves

safe and efficient driving of wheelchair. The design

method of each step is described in following

sections.

Figure 5: Corridor detection.

2.1 Corridor Detection

To achieve the two functions mentioned in section 1,

the proposed method detects the corridors. We

developed a corridor detection algorithm based on

leg detection algorithm that uses laser range finder

(LRF) (Belloto, 2009). First, it searches for the

edges as shown in Figure 5 from the LRF data using

next equation:

1ii th

ll e





(1)

Where

is the

th data of LRF and

is the

threshold. If the equation is satisfied, we assume the

th data corresponds to the edge. After edges are

detected, proposed method calculates the width

between detected edges by following equation:

res j j

wl r i i



  

(2)

Where

is the distance between the wheelchair

and

j th detected edge,

res

is the angular resolution

of LRF and





is the number of scanned point

between detected edges. Using the threshold

, if

ww

, we assumed the space between detected

edges corresponds to a corridor.

In this paper, thresholds

and

is set to 0.1 m

and 0.3 m. These are set according to the parameter

study to detect the corridors 2 m backward from it.

2.2 Evaluation of Input

To decide the moving direction roughly, the time

series of input commands is considered. In this

paper, we refer non-invasive BMIs as input device

and assume that only three different directions

(Forward, Left and Right) is fed to the wheelchair

(Millan, 2004). The operating frequency of BMI is

set to 2 Hz. The aim of this process is to select the

passenger’s intention, considering the inputting

error, such as turn and change the direction or go

straight ahead. To measure the intention we use the

following index:

() ()

Memory Memory

tnnn

Memory Memory

tnnn

RRtIRt







(3)

This index is first proposed for mobile robots to

measure the reliability of information and

()

the forgetting function based on the experimental

results on human short-term memory in psychology

field (Fujii, 2005). As this function is based on the

human’s short-term memory, we expect to achieve

intuitive driving of the wheelchair. Suffixes

L and

R indicates “Left” and “Right”. The formulation of

the forgetting function

is as follows:

() exp( )

1.0 0.3595

Rt A Bt







(4)

The

and

in equation (3) represents the input

from the device and is defined as follows:

if input is left

otherwise

if input is right

otherwise

















(5)

Therefore we can decide whether the passenger

wants to go straight ahead or turn and change

direction as the

becomes large when the

left or right side input is fed to the wheelchair

continually. In this paper, we set

as threshold

and if the index became larger than the

decide that the passenger wants to change direction.

2.3 Action Control

The action control of the wheelchair is achieved

using fuzzy potential method (FPM) (Tsuzaki,

2003). In FPM, command velocity is generated by

ICINCO 2011 - 8th International Conference on Informatics in Control, Automation and Robotics

integrating the element actions represented as

potential membership functions (PMFs). PMF is a

function that its horizontal axis represents the

traveling direction which is from -180 deg to 180

Figure 6: Example PMF for a desired direction.

Table 1: Key assignment.

Input

Forward

0.1 1.0

1.0

pifp

otherwise













Left

1tt





Right

1tt





No input

0.1 0.0

0.0

pifp

otherwise













Figure 7: Relation between wheelchair and obstacle.

deg measured from the front of wheelchair and

vertical axis represents the priority for each direction.

In this paper, three PMFs, for going to the desired

direction, to avoid obstacles and to consider

passengers intention are designed. The design

method of PMFs is described in following sections.

2.3.1 PMF for Going to the Desired

Direction

To go to the desired direction, a triangular PMF



is generated, as shown in Figure 6.



is specified by

the priority

at time t and



. These parameters

are determined by Table 1 and following equation:

0.0

cttth

if R or R R

otherwise













(6)

Figure 8: Example PMF for avoiding obstacles.

Where



is the direction of detected corridors on

the way which passenger wants to turn: i.e. if

RR the leftward and if

RR the rightward.

In case corridor isn’t detected and the equation is

satisfied,



is set to

45deg



2.3.2 PMF for Avoiding Obstacles

To avoid obstacle, PMF



is generated. First, using

the distance data from wheelchair to obstacles by

LRF, the direction not to collide to obstacles,



and



in Figure 7, considering the margin

from

the

th scanned point is calculated.





arcsin ( ) ( )

R L wheel th n

rdl





 

(7)

Next, using the calculated directions, the PMF



for

th scanned point is generated like in Figure

6. The priority

of PMF in Figure 8 is decided

using following equation.

(( ) ( ))

()

(())

1.0

nth n

nth

th th n

ldr

if l l

ldr

otherwise























(8)

Finally, the PMF



is calculated as a logical

sum of all PMF for each scanned point.

1 N

oo o

 



(9)

ACTION CONTROL METHOD FOR POWERED WHEELCHAIRS CONSIDERING CONTROL INPUT AND

ENVIRONMENTAL INFORMATION

2.3.3 PMF to Consider Passenger’s

Intention

To consider passenger’s intention, PMF



generated as shown in Figure 9. The aim of this

PMF is to restrict wheelchair from moving to the

opposite direction of passenger’s intention while

Figure 9: Example PMF to consider passenger’s intention.

avoiding the obstacles.

and

shown in Figure 9

is calculated as follows:

0.0 0

1.0

0.0 0

1.0

if R R

otherwise

if R R

otherwise





























(10)

2.3.4 Integration of PMFs and Calculation

of Command Velocity

FPM integrates the PMFs to calculate the command

velocity. The integrated PMF

mix



is calculated by

logical product of the PMFs designed in previous

sections as follows:

mix d o OR







(11)

Finally, by defuzzifier, the command velocity is

calculated as a traveling direction

out



and an

absolute value of the reference speed of the

wheelchair from the integrated PMF

mix



out



decided as the direction that makes the PMF

()

mix





maximum. Based on

out



, the traveling

speed

out

v is calculated by following equation:

max min max

()( )

tmixout

vvvv







(12)

1exp( (2 0.5))

out t

out

 







 



(13)

Where

max

v and

min

are the upper and lower limits

of the wheelchair speed. In this paper, we assume

that the wheelchair’s maximum acceleration is 0.5

m/s

and

max

v is set to 0.75 m/s while the corridors

are not detected on the left and right side of the

Figure 10: Simulation result of straight driving.

wheelchair and otherwise 0.25 m/s. 0.75 m/s is the

speed that is able to decelerate to 0.25 m/s when the

maximum acceleration of the wheelchair is 0.5 m/s

in 2 m of braking distance.

min

v is set to 0.0 m/s.

Moreover, to turn efficiently, proposed method

limits the output according to the

out



using the

sigmoid function shown in equation (13). In this

paper, the parameter



is set to



 .

3 SIMULATION

To verify the effectiveness of the proposed method,

numerical simulations were carried out. The

wheelchair size is set to L 1.05 × W 0.65 m. In order

to recognize the environment, we assumed that a

LRF is loaded on the center of wheelchair. The

measuring range of LRF is 4.0 m from -120 deg to

120 deg.

is set to 0.25 and the margin from

obstacles

is set to 0.3 m. In all simulations, the

initial position of the wheelchair is set to (0 m, 0 m).

Width of corridors is 2 m and the center of cross

road is (0.0 m, 6.0 m). The interval of grids in the

figures is 1 m. The commands of passenger are

simulated by keyboard input with operating

frequency of 2 Hz and the numbers 0 to 3 of the

input command corresponds to “0: No signal”, “1:

Forward”, “2: Left”, and “3: Right”.

3.1 Straight Driving

Figures 10 to 12 show the trajectory of wheelchair,

input commands and speed of wheelchair in straight

driving. From these results, it is verified that the

proposed method accelerates and decelerates

ICINCO 2011 - 8th International Conference on Informatics in Control, Automation and Robotics

automatically while driving the wheelchair.

Especially, while the corridor was detected,

Figure 11: The input commands in straight driving.

Figure 12: Speed of wheelchair in straight driving.

proposed method decelerates the wheelchair for safe

driving considering the undetectable moving objects

shield by the walls. By this simulation, it was

confirmed that the proposed method realizes safe

driving of wheelchair with simple input commands

in straight driving.

3.2 Direction Change

Figure 13 shows the trajectory of wheelchair with

proposed method and Figure 14 shows the trajectory

with conventional method in direction change. The

input command that was given to the wheelchair in

both methods is shown in Figure 15. Figure 16

shows the yaw angle of wheelchair in direction

change with proposed method. From the results in

Figures 13, 15 and 16, it was confirmed that

proposed method enables the passenger to change

the direction smoothly by detecting the corridors,

analyzing the input commands and directing the

wheelchair to the corridor. On the other hand, as

shown in Figure 14, as the correspondence of input

commands and the travelling direction was one-to-

one which was not suitable for turning, the

conventional method, without navigation to the

corridors, failed to change the direction and stopped.

Moreover, in case an unintended input is fed to the

wheelchair, the wheelchair with conventional

method swings as the time series of input is not

(a) Overview (b) Enlarged view

Figure 13: Simulation result in direction change with

proposed method.

(a) Overview (b) Enlarged view

Figure 14: Simulation result in direction change with

conventional method.

Figure 15: The input commands in direction change.

Figure 16: Yaw angle of wheelchair in direction change.

ACTION CONTROL METHOD FOR POWERED WHEELCHAIRS CONSIDERING CONTROL INPUT AND

ENVIRONMENTAL INFORMATION

Figure 17: Simulation result of obstacle avoidance during

straight driving.

Figure 18: The input commands while avoiding obstacles

during straight driving.

Figure 19: Speed of wheelchair while avoiding obstacles

during straight driving.

considered. Therefore it is verified that the

automatic navigation is effective for the wheelchairs

controlled by low operability devices to change

direction and it realizes smooth changing the

direction without useless stopping inside the cross

road.

3.3 Obstacle Avoidance

To verify the effectiveness of proposed method

while avoiding the obstacles, two simulations were

carried out. In the simulation two obstacles is placed

in the corridor. One is placed in (-0.7 m, 3.0 m) and

Figure 20: Simulation result of obstacle avoidance during

direction change.

Figure 21: The input commands while avoiding obstacles

during direction change.

Figure 22: Yaw angle of wheelchair while avoiding

obstacles during direction change.

the other is placed in (0.5 m, 6.0 m). The radius of

obstacle is set to 0.2 m. Figures 17 to 19 show the

trajectory of wheelchair, input commands and speed

of wheelchair in straight driving with obstacles in

the corridor. From these results, it is verified that

proposed method achieves safe driving of

wheelchair maintaining the distance from obstacles

and automatic control of velocity with simple input

commands. Especially, at 6.6 s, though the obstacle

is not placed in the travelling direction and the

passenger commands to go straight ahead, the

wheelchair automatically avoided the obstacle to

maintain the safe distance.

ICINCO 2011 - 8th International Conference on Informatics in Control, Automation and Robotics

Figure 23: Overview of wheelchair.

Figure 24: The experiment environment.

Moreover, the wheelchair analyzed the left-side

input of 20.1 s as the input is not long enough for

changing the direction. As a result, the wheelchair

avoided the obstacle from its left side instead of

changing the direction to the left.

Figures 20 to 22 show the trajectory of

wheelchair, input commands and yaw angle of

wheelchair in direction change with obstacles in the

corridor. From this simulation, by analyzing the time

series of input commands, instead of avoiding the

obstacle and going forward, the wheelchair turned

smoothly and changed its direction maintaining the

distance from obstacles. From these results, it is

verified that by analyzing the input commands and

considering the environmental information, the

proposed method achieves safe and efficient driving

of wheelchair, controlled by low operability devices,

in general corridor passing situations without

complex commands.

4 EXPERIMENT

4.1 Experiment Environment

To verify the effectiveness of proposed method in

real environment, experiment with real wheelchair

was carried out. The wheelchair is YAMAHA JWX-

1 with a HOKUYO URG04-LX LRF on the front of

wheelchair to detect corridors and obstacles. Figure

23 shows the overview of wheelchair.

Figure 25: The input commands during the experiment.

Figure 26: Experiment result of direction change (back).

Figure 27: Experiment result of direction change (side).

LRF used in the experiment has 4 m of measurement

range from 120 deg to -120 deg. The control of

wheelchair is achieved with a laptop PC with

Centrino-Duo CPU 2.0 GHz. The size of wheelchair

is L 1.05×W 0.65 m. The maximum speed of

wheelchair is set to 0.5 m/s. The input commands

are given to the wheelchair by keyboard manually

with operating frequency of 2 Hz. Figure 24 shows

the experiment environment.

4.2 Experiment Result

Figure 25 shows the input command that was given

to the wheelchair during the experiment and Figures

26 and 27 show the experiment result from different

view angles. From these results, it was confirmed

ACTION CONTROL METHOD FOR POWERED WHEELCHAIRS CONSIDERING CONTROL INPUT AND

ENVIRONMENTAL INFORMATION

that proposed method automatically analyze the

passenger’s objective and changes the wheelchair’s

direction without complex commands. Moreover, it

was verified that the proposed method achieves safe

and smooth driving of wheelchair also in real

environment by considering the passenger’s control

input and environmental information simultaneously

while driving through the corridors.

5 CONCLUSIONS

In this paper, to control the wheelchair with devices

alternative to joystick safely and efficiently, a novel

action control method for powered wheelchairs

controlled by low operability device, low operating

frequency and few input direction, is proposed.

In order to achieve safe and efficient action while

driving the corridors, the proposed method considers

the time series of input commands and

environmental information simultaneously.

Moreover, to detect the corridors, a corridor

detection algorithm is also proposed. Through the

numerical simulation and experiment, efficient and

safe driving of wheelchair, efficient turns to change

the direction in cross roads and safe driving with

automatic control of velocity was achieved by

proposed method and its effectiveness was verified.

REFERENCES

Simpson, R., LoPresti E., Cooper R., 2008. How many

people would benefit from a smart wheelchair?.

Journal of Rehabilitation Research & Development.

Levine, S., Bell D., Jaros L., Simpson R., Koren Y.,

Borenstein J., 1999. The NavChair Assistive

Wheelchair Navigation System. IEEE Transaction on

Rehabilitation Engineering, vol. 7, No. 4, pp. 443-451.

Simpson, R. and Levine S., 2002. Voice Control of a

Powered Wheelchair. IEEE Trans. on Neural Systems

and Rehabilitation Engineering, vol. 10, No. 2, pp.

122-125.

Saitoh, T., Takahashi N., Konishi R., 2007. Development

of an Intelligent Wheelchair with Visual Oral Motion.

IEEE International Conference on Robot & Human

Interactive Communication, pp. 145-150.

Vanacker, G., Millan J., Lew E., Ferrez P., Moles F.,

Philips J., Brussel H., Nuttin M., 2007. Context-Based

Filtering for Assisted Brain-Actuated Wheelchair

Driving. Computational Intelligence and Neuroscience

Volume 2007, pp. 1-12.

Iturrate, I., Antelis J., Kubler A., Minguez J., 2009. A

Noninvasive Brain-Actuated Wheelchair Based on a

P300 Neurophysiological Protocol and Automated

Navigation. IEEE Transaction on Robotics, Vol. 25,

No. 3, pp. 614-627.

Millan J., Renkens F., Mourino J., Gerstner W., 2004.

Noninvasive Brain-Actuated Control of a Mobile

Robot by Human EEG. IEEE Transaction on

Biomediacl Engineering, Vol. 51, No. 6, pp. 1026-

1033

Fujii, H. and Yoshida K., 2005. Action Control Method

for Mobile Robot Considering Uncertainty of

Information. IEEE/RSJ International Conference on

Intelligent Robots and Systems, pp. 3915-3920.

Belloto, N. and Hu H., 2009. Multisensor-Based Human

Detection and Tracking for Mobile Service Robots.

IEEE Transaction on Systems, Man, and Cybernetics-

Part B: Cybernetics, Vol. 39, No. 1, pp. 167-181.

Tsuzaki, R. and Yoshida K., 2003. Motion Control Based

on Fuzzy Potential Method for Autonomous Mobile

Robot with Omnidirectional Vision. Journal of the

Robotics Society of Japan, Vol. 21, No. 6, pp. 656-662.

ICINCO 2011 - 8th International Conference on Informatics in Control, Automation and Robotics