A Study on Gathering Staircase Information for Active Staircase

Entry of Wheelchair Stair Climbing Assistive Devices

Su-Hong Eom

, Jeon-Min Kang

, Ga-Young Kim

and Eung-Hyuk Lee

Department of Electronic Engineering, Tech University of Korea, Siheung City, Gyeonggi-do, South Korea

Keywords: Stairs Information, Autonomous Driving on Stairs, Intelligent Wheelchairs, LiDAR.

Abstract: Wheelchairs are the most commonly used auxiliary devices by people with mobility impairments, and

autonomous driving technology has recently been applied to these wheelchairs using robot technology.

However, in an autonomous driving environment, most stairs are recognized as obstacles. For autonomous

driving on stairs, recognition of stairs information must be preceded. Currently, the classification of stairs into

stairs and non-stairs is performed based on vision sensors and can be determined by a high recognition rate.

However, the measurement and estimation of the riser height value, tread depth value, and angle of pitch

value of stairs are not. Therefore, this study proposes a method of obtaining the shape information of stairs

using 2D LiDAR. The proposed method measured the riser height and tread depth of stairs using the K-Means

and RANSAC algorithm after obtaining the raw data by rotating the 2D LiDAR by 90 degrees, and based on

this, the angle of pitch value was calculated. The riser height and tread depth values were determined by about

±13mm on average, and the angle of pitch value showed the accuracy of ±1° accuracy through applying a

quantitative verification method for the proposed method.

1 INTRODUCTION

It is paying attention to the rapid increase in the

elderly group (aged people) since COVID-19

throughout the world. According to the "World Social

Report 2023" published by the UN, the number of

elderly people aged 65 or older is expected to reach

1.6 billion by 2050 (United Nations, 2023). It is

analyzed that this increase trend is progressing faster

in developing countries than in developed countries.

Such elderly people have difficulty walking due

to physical aging, and the frequency of outdoor

activities decreases compared to the younger age,

making it worse as they have disabilities (Miodrag

Počuč et al., 2021). These activity constraints become

passive in voluntary social participation and can lead

to psychological depression or lack of self-esteem

(AH Taylor et al., 2004).

People who have difficulty walking due to

physical ageing and disability are referred to as

mobility impairments, and most of them use cars with

https://orcid.org/0000-0001-8493-1432

https://orcid.org/0009-0008-9543-3373

https://orcid.org/0000-0003-4113-5457

https://orcid.org/0000-0002-4434-0694

guardians for convenience of their movement

(Miodrag Počuč et al., 2021; United Nations, 2022).

However, this situation suggests that more caregivers

are needed in the future society, but it is currently

expected that supply versus demand will not be kept

up due to a severe drop in fertility rates (United

Nations, 2023).

Currently, the elderly and the disabled people,

who are facing mobility impairments, use

wheelchairs as a means of mobility assistance after

cars, but they do not meet their mobility

independence. Therefore, the intelligence of

wheelchairs using robot technology is being actively

studied.

In the past, the study of intelligent wheelchairs

was mainly aimed at manipulating interfaces, posture

change, and obstacle recognition and avoidance for

the purpose of use by various disabled groups (Jesse

Leaman et al., 2017; Amiel Hartman et al., 2019).

In recent research on intelligent wheelchairs,

autonomous driving technology using robot

Eom, S., Kang, J., Kim, G. and Lee, E.

A Study on Gathering Staircase Information for Active Staircase Entry of Wheelchair Stair Climbing Assistive Devices.

DOI: 10.5220/0012255700003543

In Proceedings of the 20th International Conference on Informatics in Control, Automation and Robotics (ICINCO 2023) - Volume 1, pages 747-753

ISBN: 978-989-758-670-5; ISSN: 2184-2809

747

technology has become a hot topic to solve the social

problems raised above. Since wheelchairs are not

much different from mobile robots due to their

mechanical characteristics, studies based on the AMR

(Autonomous Mobile Robot) technology are being

attempted (André R. Baltazar et al., 2019). However,

since the use environment of wheelchairs is used in

daily life, unlike AMR, further studies on the

detection and avoidance of various obstacles are

needed. Among them, countermeasures against stairs

are a must-resolve challenge. The reason for this is

that wheelchair users refer the environment as the

biggest travel restriction in their daily lives as mound

and stairs driving (Korean Consumer Agency, 2011).

Currently, the platform for wheelchairs to move

on stairs adopts an orbital wheel structure, but the

technology to automatically recognize and move on

stairs is insignificant (Weijun Tao et al., 2017; Bibhu

Sharma et al., 2022). A vision-based study is a

representative technology for recognizing stairs.

However, these technologies were developed for the

purpose of AMR or mobile robots to recognize stairs

as obstacles rather than driving environments, and the

problems and approaches raised in this study are

different (M Basavanna et al., 2021).

With the development of vision-based AI

technology after the 4th industry, the recognition rate

of stairs environments is more than 90% based on the

technology of image matching and pattern

comparison, but the rate in this technology for

estimating the stairs shape information, riser height

and tread depth, which are necessary for moving on

stairs, is not high (Chen Wang et al., 2023).

The reason why such information is needed is that

the angle of pitch value of the pitch line derived from

the shape of stairs is required for the orbital wheel

platform to ensure safe driving on stairs. If this

information is not recognized, a large impact occurs

due to the change in the center of gravity of the

platform in the landing section at the beginning and

end of stairs (Daisuke Endo et al., 2017).

The following two types of research on the stairs

information estimation are representative. The first is

a method of estimating stairs based on a number of

single distance detection sensors, and the recognition

rate is not high in the form of estimating the

approximate height of stairs and inferring the pitch

line, and it may not be applicable depending on the

stairs driving platform (Su-Hong Eom et al., 2020;

Hyun-Chang Hwang et al., 2021).

The second method uses a depth camera, which

has an excellent effect on straight line detection of

stairs through an algorithm such as the Hough

transform method for edge detection after

preprocessing image information, but has a problem

of varying accuracy due to a relatively low

recognition rate and some external environmental

factors such as light exposure (Jia sheng Liu et al.,

2020; Haruka Matsumura et al., 2022).

Therefore, this study proposes a method of

estimating the riser height and tread depth of stairs,

and the angle of pitch value of the pitch line using 2D

LiDAR in order to solve the problems of the existing

method.

Inferring the information of stairs based on the

information from LiDAR causes an increase in the

cost of the system compared to the method mentioned

above. However, the purpose of this study is that

wheelchair users autonomously drive on stairs using

an infinite orbit platform, and it is assume that the

platform is already equipped with a LiDAR sensor

system.

The proposed method is a little more intuitive

when using 3D LiDAR, but in an autonomous driving

system in mobile robots, 2D LiDAR is generally used

to solve the increase in the cost of operating its

system. For this reason, the stairs were vertically

scanned by rotating the LiDAR by 90° to estimate the

stairs information.

2 METHOD

The riser height and tread depth of stairs vary

depending on the manufacturing and installation

environment, and the angle of pitch value of the

general walkable stairs is about 45° (HM

Government, 2013). as shown in Figure 1. However,

the mechanical characteristics of stairs are the same.

In addition, the shape measured may differ depending

on the operating principle or operation method of the

applied sensor. In this study, since 2D LiDAR is

applied to vertical plane measurement rather than

horizontal plane measurement, it is necessary to

convert the measurement data into the same form as

seen with the human eye through a coordinate

transformation process, and to adjust the sensing

range to be measured.

Figure 1: Schematic of stairs measurement limitations using

LiDAR.

ICINCO 2023 - 20th International Conference on Informatics in Control, Automation and Robotics

748

The riser height and tread depth measurements of

stairs require the separation of data for measuring

stairs information from the LiDAR raw data and the

data unnecessary for measurement. This is because

the tread depth cannot be measured depending on the

location of stairs viewed from LiDAR. Therefore, this

section describes the method of calculating the angle

of pitch value based on the riser height and tread

depth of stairs together with the above-mentioned

parts.

In this paper, the wheelchair direction is

straightforward to the staircase to scan the stairs

vertically with 2D LiDAR, and only the ascending

staircase is considered.

2.1 Limiting the Scan Area of the

LiDAR Sensor and Transforming

the Plane Coordinate System

2.1.1 Limiting the Scan Area of the Lidar

Sensor

LiDAR has a different scan range depending on the

product specifications. Therefore, it is necessary to

limit the scope to measure stairs at an installation

location, otherwise more information is measured

together, which acts as a noise component. For

example, in the case of the measurement situation as

shown in Figure 1, the environmental information

with the stairs is measured together, and there is a

trouble of performing an additional preprocessing

process to estimate the stairs information.

Therefore, this study attempts to limit the scan area

when measuring by LiDAR to prevent such problems

from occurring.

In this paper, the LiDAR scan area aims to set a

scan range of 15° upward (left) and 40° downward

(right) from the center of the scan range. This may

vary depending on the location of the LiDAR

installation. The effect of this setting is shown in

Figure 2.

Figure 2: LiDAR raw data before and after the scan range

limit.

2.1.2 Coordinate Transformation of the

LiDAR Sensor

LiDAR raw data is generally output as an angle value

and a distance value located on the target object

around the LiDAR. As these data are polar

coordinate, it is necessary to transform it into a two-

dimensional planar coordinate system in order to

estimate information on stairs. This coordinate

system transformation is performed by Equation (1)

and (2). In Equation (1) and (2), the negative

multiplication is based on the direction of the LiDAR

installation and may be omitted depending on the

situation.

𝑥



1



𝛾cos 𝜃 (1)

𝑦



1



𝛾𝑠𝑖𝑛 𝜃 (2)

2.2 Preprocessing Methods for

Measuring Stairs Information from

LiDAR Data

2.2.1 Data Clustering

Since the object information data of LiDAR consists

of distance information and angle information by

irradiating a laser in a fan shape at the center of

LiDAR, there are measurable and unmeasurable parts

depending on the location of the object and LiDAR.

In a data collection environment as shown in Figure

1, the first tread depth of the stairs may be

measurable, but depending on the height of the

LiDAR position, the tread depth of the second step

may not be measured. Therefore, it is necessary to

separate segments for measuring stairs information

from the LiDAR raw data.

The segment separation means the separation of

valid data and ineffective data. This process is

possible using a machine learning method. In the

machine learning method, the data separation is

performed as a way of classification or clustering, and

SVM (Support Vector Machine), KNN (K-Nearest

Neighbors), K-Means, and Hierarchical Clustering

are representative (Shweta Mittal et al., 2019;

Abiodun M. Ikotun et al., 2022). Among them, this

study aims to perform clustering through the K-

Means algorithm because the purpose is to implement

clustering based on the location similarity of data.

The reason for choosing clustering by the K-

Means algorithm is that the hierarchical clustering

algorithm goes through a hierarchical clustering

process and has the disadvantage of considering the

distance and similarity of cluster data in advance. In

addition, this is because the computational

A Study on Gathering Staircase Information for Active Staircase Entry of Wheelchair Stair Climbing Assistive Devices

749

complexity is higher than that of the K-Means

algorithm in real-time reflection of this system.

Although it cannot be concluded that the K-Means

algorithm has a shorter calculation time than the

Hierarchical Clustering algorithm, the method

proposed in this study is possible because the area

was limited when collecting the LiDAR raw data.

2.2.2 Data Linearization

In calculating the riser height and tread depth length

of the stairs from the clustered data, it is necessary to

linearize the data in order to calculate the angle of

pitch of the stairs from the clustered data showed in

2.2.1. This is because the LiDAR raw data results in

shape errors caused by foreign substances present in

the stairs and errors due to damages of the edge of the

stairs. the data linearization can be performed by

simply using a moving average filter, but it causes

data errors when unexpected noise components are

introduced. Therefore, in this study, the RANSAC

(Random Sample Consensus) algorithm is used. The

LSM (Least Square Method) algorithm is also

applicable, but the RANSAC algorithm is effective in

this case because it needs to be linearized into a

straight component of non-damaged stairs from the

previously mentioned data such as edge breakage

(Sunglok Choi et al., 2009).

2.2.3 Stairs Entry Angle Calculation

Algorithm for Stable Stairs Driving of

Orbital Wheels

As shown in Figure 1, the pitch line of the stairs

connects the end points of the stairs tread and

becomes a driving path when the track-type wheel

drives on the stairs. Here, the angle of the driving path

can be calculated as a trigonometric function with the

riser height value and the tread depth value as shown

in Equation (3), (4), and (5). The riser height value

selects a cluster with a large area value among

clusters derived through the K-Means algorithm,

linearizes the selected cluster through the RANSAC

algorithm, and segments continuous Y-axis data.

Based on the segmented data, the coordinates of both

end points of riser-1 and riser-2 are obtained as shown

in Figure 3, and the tread depth is calculated based on

this. Figure 4 shows its schematic diagram.

stair depth 

𝑃



x𝑃



(3)

stair height 

𝑃



𝑃



(4)

stair angle  tan





stair height

stair width



(5)

Figure 3: Schematic diagram for calculating the entrance

angle of stairs.

Figure 4: Block diagram for calculating the entrance angle

of stairs.

3 EXPERIMENT AND RESULTS

3.1 Experimental Environment

For the theoretical verification of the method

presented in this study, an experiment was conducted

in a stairs environment as shown in Figure 5. The

experiment was conducted by rotating the 2D

LiDAR, which can measure the X-axis in a two-

dimensional plane at a height of 55cm from the

bottom of the stairs and 45cm from the front, by

limiting the angle of the Y-axis to a scan angle of

+15° to -40° based on 0°.

ICINCO 2023 - 20th International Conference on Informatics in Control, Automation and Robotics

750

Table 1: Stairs specification.

Stair riser height 180mm

Stair tread depth 260mm

Stair angle of pitch 35°

Table 2: LiDAR sensor specification.

Manufacturer /

Product Name

SLAMTEC /

RPLiDAR S1

Scan rate 10Hz

Scan angle 180°

Angular Resolution 0.35°

Max Distance range

white object 40m, black

object 10m

Figure 5: Experimental environment.

3.2 Experimental Results

3.2.1 Stairs Raw Data

For the theoretical verification of the method

presented in this study, the stairs were measured

as shown in Figure 1 in the experimental environment

as presented in Figure 5. As a result, it was detected

in three areas in the LiDAR scan area as shown in

Figure 6.

Shape-1 means the shape of the first step, and

Shape-2 and Shape-3 mean the riser of the second and

third steps respectively. The reason why the tread was

not measured from the second step is because of the

LiDAR measurement height, as described in 2.1.

Figure 6: Scan results of the stairs using 2D LiDAR.

3.2.2 Stairs Raw Data

For clipping Shape-1 only from the measured raw

data of the stairs (Figure 6), the clustering was

performed using the K-Means algorithm, and the

number of clustering for the cluster was determined

by two. The initial value of the centroid was set to a

random value and the K-Means++ method was

applied. The result is shown in Figure 7, and the

Figure 7: Results of the stairs data clustering using the

K-means and its centroid.

Figure 8: Measurement target stairs selected as a result of

comparison between each cluster.

A Study on Gathering Staircase Information for Active Staircase Entry of Wheelchair Stair Climbing Assistive Devices

751

centroid is a star shape in the graph. For the cluster

clipping for an actual stair information measurement

in two clusters, the clipping was performed through

comparing the area size of the cluster, and the results

are shown in Figure 8.

Algorithm 1 shows the pseudo code implemented

to segment the continuous Y-axis data for detecting

risers in this clipped Shape-1 data, and the set by user

value in the case of the set distance during this

experiment is 0.007 and the minimum segment length

is 10.

Input: set distance = set by user, index number =

0, minimum segment length = set by user

Result: segmented lines

for i from start to end of LiDAR data do

if calculate distance(x[i], y[i], x[i+1], y[i+1])

> set distance then

index number = i + 1

if i – index number > minimum segment

length

coordinates of x = from index number to i

in LiDAR data of x coordinate

coordinates of y = from index number to i

in LiDAR data of y coordinate

else

pass

else

pass

Algorithm 1: Segment separation algorithm for detecting

risers.

Figure 9 shows the results of applying the

RANSAC algorithm to the detected segments based

on the applied algorithm, and based on these two riser

coordinate values, the riser height and tread depth

were measured using Equation (3), (4) and (5) and the

angle of pitch value was calculated.

Figure 9: Results of applying RANSAC for the riser height.

The angle of pitch value is shown in Table 3.

These values are the results of 10 repeated

measurements while maintaining a certain distance in

the environment presented in Figure 5. The

measurement results showed that the riser height and

tread depth values were around ±13mm on average

and the angle of pitch value was ±1° accuracy

compared to the actual stairs information.

Table 3: Segment separation algorithm for detecting risers.

Trials

Riser

height[m]

Tread

depth[m]

Angle of

pitch[°]

1 0.272 0.186 34.3

2 0.269 0.198 36.3

3 0.271 0.186 34.4

4 0.27 0.198 36.2

5 0.268 0.186 34.7

6 0.269 0.199 36.4

7 0.269 0.187 34.8

8 0.271 0.198 36.1

9 0.271 0.198 36.1

10 0.269 0.198 36.3

Avg. 0.269 0.193 35.6

4 CONCLUSIONS

This study presented a method for calculating the

angle of pitch value based on shapes of stairs in order

to actively drive on stairs when entering or exit the

stairs using a wheelchair or wheelchair-combined

auxiliary device using an orbital wheel platform. The

proposed method used the K-Means and RANSAC

algorithms as preprocessing algorithms to vertically

scan the stairs using a 2D LiDAR sensor and to

measure the riser height and tread depth of the stairs

based on the distance and angle values between the

sensor and the stairs. The riser height value and tread

depth value were measured based on the two riser

coordinate values finally derived through the

RANSAC algorithm. Finally, the angle of pitch value

of the stairs was calculated using a trigonometric

function based on the measured riser height value and

tread depth value. The calculation results showed an

average of 36° in 10 trials, which showed an average

error of about 1° compared to the actual step

information, verifying the appropriateness of the

proposed algorithm.

Based on the algorithm presented in this study, it

is expected that it will be more flexible and safe drive

on stairs when entering or exit the stairs if applied as

a control parameter for autonomous stairs driving of

a wheelchair or wheelchair-combined assistive

ICINCO 2023 - 20th International Conference on Informatics in Control, Automation and Robotics

752

device using an orbital wheel platform. However, the

algorithm presented in this study is a method verified

at 45cm from the front of the stairs and 55cm in

height, and the sensor attachment location should be

selected in consideration of this.

ACKNOWLEDGEMENTS

This research was supported by the MSIT(Ministry of

Science and ICT), Korea, under the ITRC

(Information Technology Research Center) support

program (IITP-2023-2018-0-01426) supervised by

the IITP(Institute for Information & Communications

Technology Planning & Evaluation)

This research was supported by 2023 Regional

Industry-linked University Open-Lab Development

Support Program through the Commercializations

Promotion Agency for R&D Outcomes (COMPA)

funded by Ministry of Science and ICT

(2023openlab(RnD)_01)

This research was supported by a grant of the

Korea Health Technology R&D Project through the

Korea Health Industry Development Institute

(KHIDI), funded by the Ministry of Health & Welfare,

Republic of Korea (HJ20C0058).

REFERENCES

United Nations. (2023). World Social Report 2023:

Leaving No One Behind In An Ageing World.

Miodrag Počuč, Valentina Mirović, Jelena Mitrović Simić,

Caglar Karamasa. (2021). Mobility analysis of persons

with disabilities. Discrete Dynamics in Nature and

Society.

AH Taylor, NT Cable, G Faulkner, M Hillsdon, M Narici,

AK Van Der Bij. (2004). Physical activity and older

adults: a review of health benefits and the effectiveness

of interventions. Journal of Sports Sciences. vol. 22, no.

8, pp. 703-725.

United Nations. (2022). Disability-Inclusive Communica-

tions Guidelines.

Jesse Leaman, Hung Manh La. (2017). A Comprehensive

Review of Smart Wheelchairs: Past, Present, and

Future. IEEE Transactions on Human-Machine

Systems, vol. 47, no. 4, pp. 486-499.

Amiel Hartman, Vidya K. Nandikolla. (2019). Human-

Machine Interface for a Smart Wheelchair. Journal of

Robotics.

André R. Baltazar, Marcelo R. Petry, Manuel F. Silva,

António Paulo Moreira. (2021). Autonomous

wheelchair for patient’s transportation on healthcare

institutions. SN Applied Sciences. vol. 3, no. 3.

Korean Consumer Agency. (2011). A Survey on Safety

Accidents in Wheelchairs.

Weijun Tao, Junyi Xu, Tao Liu. (2017). Electric-powered

wheelchair with stair-climbing ability. International

Journal of Advanced Robotic Systems. vol. 14, no. 4.

Bibhu Sharma, Branesh M. Pillai, Korn Borvorntanajanya,

Jackrit Suthakorn. (2022). Modeling and Design of a

Stair Climbing Wheelchair with Pose Estimation and

Adjustment. Journal of Intelligent & Robotic Systems.

vol. 106.

M Basavanna, M Shivakumar, K.R Prakash, Pratham

Bhomkar. (2021). ROS Based 3D Mapping of an

Indoor Environment Using Fusion of Orbbec Astra

Camera and Lidar on Turtlebot Mobile Robot. 2021 5th

International Conference on Electrical, Electronics,

Communication, Computer Technologies and

Optimization Techniques (ICEECCOT). pp. 323-327.

Chen Wang, Zhongcai Pei, Shuang Qiu, Zhiyong Tang.

(2023). RGB-D-Based Stair Detection and Estimation

Using Deep Learning. Sensors. vol. 23, no. 4.

Daisuke Endo, Atsushi Watanabe, Keiji Nagatani. (2017).

Stair Climbing Control for 4-DOF Tracked Vehicle

Based on Internal Sensors. Journal of Robotics.

Su-Hong Eom, Sun-Jong Na, Jung-Hwun You, Hyun-

Chang Hwang, Eung-Hyuk Lee. (2020). A Study on

Estimation of a Stair Entry Angle and Operation for The

Stair Climbing Aid Platform of Wheelchair. 2020

International Conference on Electronics, Information,

and Communication (ICEIC). pp. 867-869.

Hyun-Chang Hwang, Won-Young Lee, Jong-Hee Ha,

Eung-Hyuck Lee. (2021). A Study on Autonomous

Stair-climbing System Using Landing Gear for Stair-

climbing Robot. Journal of IKEEE. vol. 25, pp. 362-

370.

Jia sheng Liu, Jian po Guo, Zhen kai Xiong, Hua Li. (2020).

Design of intelligent recognition and positioning

algorithm for stairs based on depth camera. 2020 2nd

International Conference on Machine Learning, Big

Data and Business Intelligence (MLBDBI). pp. 339-

343.

Haruka Matsumura, Chinthaka Premachandra. (2022).

Deep-Learning-Based Stair Detection Using 3D Point

Cloud Data for Preventing Walking Accidents of the

Visually Impaired. IEEE Access. vol. 10, pp. 56249-

56255.

HM Government. (2013). Protection from falling, collision

and impact.

Shweta Mittal, Om Prakash Sangwan. (2019). Big Data

Analytics using Machine Learning Techniques. 2019

9th International Conference on Cloud Computing,

Data Science & Engineering (Confluence). pp. 203-

207.

Abiodun M. Ikotun, Absalom E. Ezugwu, Laith Abualigah,

Belal Abuhaija h, Jia Heming. (2022). K-means

clustering algorithms: A comprehensive review,

variants analysis, and advances in the era of big data.

Information Sciences. vol. 622, pp. 178-210.

Sunglok Choi, Taemin Kim, Wonpil Yu. (2009). Robust

Video Stabilization to Outlier Motion using Adaptive

RANSAC. 2009 IEEE/RSJ International Conference

on Intelligent Robots and Systems. pp. 1897-1902.

A Study on Gathering Staircase Information for Active Staircase Entry of Wheelchair Stair Climbing Assistive Devices

753