Development of Cart with Providing Constant Steerability Regardless

of Loading Weight or Position: 3

Report on Evaluation of a Steering

Assist System on Translational Movement

Shunya Aoki

, Sho Yokota

, Akihiro Matsumoto

, Daisuke Chugo

, Satoshi Muramatsu

Katsuhiko Inagaki

and Hiroshi Hashimoto

Dept. of Mechanical Engineering, Toyo University, Saitama, Japan

School of Engineering, Kwansei Gakuin University, Sanda, Japan

Dept. of Applied Computer Eng., Tokai University, Hiratsuka, Japan

Adv. Institute of Industrial Tech, Shinagawa, Japan

Keywords: Mechatronics, Sensing, Active Steering Caster, Shopping Cart, Passive Robotics.

Abstract: The steering of the shopping cart is affected by its load, leading to the need for the user to make corrective

adjustments and apply excessive force. In this study, the system assists the steering of a shopping cart by

actively steering casters based on the user’s operational intention estimated from the user’s force. This paper

provides a brief introduction to the operational interface and the active steering caster. Subsequently, it

elaborates on the steering assist system designed for translational movement. Furthermore, we conduct

experiments to evaluate the steerability through subjective and objective assessments. These results confirmed

that the system can support the operating force and corrective steering. In addition, subjects feel less weight

than the conventional carts and have more intuitive sense than the conventional cart.

1 INTRODUCTION

The shopping cart is designed to carry heavy or many

items while providing intuitive omnidirectional

movement. These advantages enhance the overall

comfort of shopping experiences for customers

(hereafter referred as “users”). However, the

steerability of a shopping cart is influenced by several

factors such as the weight and position of loads or

items. Consequently, users are often compelled to

make corrective steering or exert excessive force

when operating the cart. As a result, this situation can

lead to discomfort and decreased safety. Addressing

these issues can be achieved by ensuring constant

steerability of the shopping cart, irrespective of the

weight or position of the load. Therefore, the primary

objective of this study is to develop a shopping cart

that provides constant steerability regardless of the

loading weight and position.

https://orcid.org/0000-0002-8507-5620

https://orcid.org/0000-0002-3004-7235

https://orcid.org/0000-0002-3884-3746

https://orcid.org/0000-0003-2416-8038

Several studies have been conducted on assisting

cart operation, but there are some particular

challenges and limitations. For instance, in the case

of the ‘A Person-Following Shopping Supported

Robot (Islam, 2019)’, the robot tracks the user’s

position using omnidirectional camera and ultrasonic

sensor. However, it has been observed that the

absence of physical contact between the user and the

cart results in decreased sense of agency (Yun, 2017),

potentially increasing the risk of accidents such as

collisions in a store. In case of the ‘Omnidirectional

Power-assisted Cart (Maeda, 2003)’, the cart achieves

omnidirectional movement using omni-wheels to

align with the user’s intention. However, due to the

motor-generated driving force, the system poses an

increased risk of the shopping cart losing control.

Another problem arises when the cart remains

immobile when assistance is not needed. Such

problems also occur in another study, ‘Add-on Type

Aoki, S., Yokota, S., Matsumoto, A., Chugo, D., Muramatsu, S., Inagaki, K. and Hashimoto, H.

Development of Cart with Providing Constant Steerability Regardless of Loading Weight or Position: 3 rd Report on Evaluation of a Steering Assist System on Translational Movement.

DOI: 10.5220/0012200300003543

In Proceedings of the 20th International Conference on Informatics in Control, Automation and Robotics (ICINCO 2023) - Volume 1, pages 689-696

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

689

Figure 1: Conceptual illustration of the steering-assisting

cart.

Electric Wheelchair Using Active Caster (Nakayama,

2023)’. Another limitation observed in studies such

as Passive Intelligent Walker (Hirata, 2007) is an

inability to achieve omnidirectional movement due to

the fixed casters.

To address these challenges, we propose steering-

assisting carts that incorporate passive robotics, as

illustrated in Figure 1. This system utilizes the

operational interface (Aoki, 2023), where strain

gauges are directly bonded to the handle posts, to

estimate the user's operational intention. The active

steering casters (Aoki, 2022) generate the necessary

steering force to assist in the cart's operation.

Essentially the moving direction of a cart with all

casters hypothetically fixed is restricted to one

direction. Thus the steering angle of the casters

determines the moving direction of the cart. In the

proposed system, the actuator actively steers the

caster in intended direction by user, allowing the user

to advance the cart in the intended direction simply

by applying a pushing force. This constant

steerability makes shopping more comfortable and

safer. In addition to improving the shopping

experience, this system can ease the work in a

warehouse or any loading station. Figure 2 illustrates

the prototype assembled to validate the function of

the individual components. Chapter 2 provides a

concise overview of these components.

In this paper, we develop a steering-assisting

system designed explicitly for translatory movement.

Specifically, we discuss how the system estimates a

user’s intended translatory moving direction and

accordingly controls the active steering casters to

follow this user intention. Furthermore, we

comprehensively evaluate the steerability using both

subjective and objective assessments. To be more

specific, we compare the active steering cart with

passive steering cart in terms of the steering and

corrective force required by the user and their

subjective experience of operability.

Figure 2: Prototype of the steering-assisting cart.

2 SYSTEM CONFIGURATIONS

This chapter briefly overviews the operational

interface (Aoki, 2023), the active steering caster

(Aoki, 2022), and its wheel arrangement in order to

enhance the paper's readability.

2.1 Operational Interface

The operational interface serves as a sensing system

to estimate the user's operational intention. In this

study, we propose an operational interface that

measures the user's operating force 𝐹



,𝐹



 [N] and

moment of force 𝑀



[Nm]. The intention is estimated

using a method based on the omnidirectional power-

assisted cart (Maeda, 2003) as a reference. While

various methods have been proposed for such

operational interfaces, it is essential to note that they

also come with their challenges. For example, some

operational interfaces face challenges such as

complex mechanisms by using a 6 DOF force-torque

sensor (Ueno, 2014), the need for mounting movable

parts by using a potentiometer (Seino, 2017), and the

requirement to measure objects other than the cart's

frame by using an indirect displacement sensor

(Maeda, 2003). To address these challenges, we

propose an operational interface in which strain

gauges are directly bonded to the handle posts. This

interface utilizes the structure's strain as the

operational interface's input. Consequently, the

system eliminates the need for additional parts,

allowing for the direct measurement of the operating

force. Figure 3 showcases the prototype of the

operational interface described above. The strains on

the handle posts 𝜀



,𝜀



,𝜀







are converted into the

applied forces on the handle posts using equation (1).

Furthermore, these forces are then transformed into

the operating force 𝐹



,𝐹



[N] and moment of force

𝑀



[Nm] on the cart using equation (2).

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

690

Figure 3: Prototype of the operational interface and

operating force conversion flow.



𝐹



𝐹



𝐹







𝑨



𝜀



𝜀



𝜀



𝑩 (1)

here

𝑨





𝛼



0𝛼



00𝛼





,𝑩

𝑒



𝑒



𝑒



 .



𝐹



𝐹



𝑀





⎝

⎜

⎛

001

𝑑



𝑑

⎠

⎟

⎞



𝐹



𝐹



𝐹



 (2)

In this context, the matrices A and B represent the

coefficient matrix and the offset matrix, respectively,

which facilitate the conversion of the strains

𝜀



,𝜀



,𝜀







into the corresponding force on the

handle posts 𝐹



,𝐹



,𝐹



[N]. The parameter 𝑑 [m]

denotes the distance between the handle posts.

2.2 Active Steering Caster

The active steering caster plays a crucial role as a

fundamental component of the steering-assisting cart.

It generates the steering force and determines the

arbitrary moving direction of the cart. Figure 4 shows

the prototype of the active steering caster. It

incorporates a stepping motor to drive the steering

axis and an electromagnetic clutch for switching

between active and passive steering modes. This

design serves as a backup system, allowing the caster

to function as a conventional cart that passively steers

when assistance is unnecessary or when the battery is

depleted. The problems with related research are

resolved by the driven steering axis (not driven wheel

axis) and the switching mechanism.

Figure 4: Prototype of the active steering caster.

The steering angle of the active steering caster

determines the traveling direction of the cart. To

achieve determine the arbitrary moving directions for

the cart, it is necessary to arrange two active steering

casters among the four casters and control their

respective steering angles. This configuration allows

for control over the cart's movement in various

directions. However, when the wheel axes of the

active steering casters are aligned in a straight line, as

depicted in Figure 5, the center of rotation can be any

point along this line. In other words, when the wheel

axes of the active steering casters are aligned on the

same straight line, the traveling direction is not

uniquely determined. We proposed that two active

steering casters and two conventional casters are

arranged diagonally to reduce this problem (Aoki,

2022).

Figure 5: Indeterminate traveling direction with two active

steering caster axes on the same straight line.

3 STEERING-ASSISTING

SYSTEM

The proposed system is implemented by estimating

the operational intention and commanding the

steering angle for the active steering casters, as well

as other functions. The communication between

Development of Cart with Providing Constant Steerability Regardless of Loading Weight or Position: 3 rd Report on Evaluation of a

Steering Assist System on Translational Movement

691

processes in the system is facilitated by the Robot

Operating System (ROS). The ROS network, which

visualizes the interconnectedness of these processes,

is depicted in Figure 6. The steering-assisting system

is implemented by establishing a unidirectional data

flow of operating force information among the ROS

nodes. In this chapter, we provide detailed

explanations regarding the role, algorithm, and

specifications of the involved hardware in the system.

Additionally, Table 1 presents a comprehensive

overview of the hardware components and their

corresponding elements utilized in this system.

The "/interface_node" measures the strains on the

handle posts. The strains on the handle posts, induced

by the operating force, change the resistance of the

strain gauges. The strain amplifiers are utilized to

convert the resistance variation into a corresponding

voltage variation. However, it is essential to note that

the strain amplifiers' output may contain some noise.

To mitigate the effects of noise, a low-pass filter with

a cut-off frequency of 𝑓



1.59[Hz] is employed to

smooth the output of the strain amplifiers.

Subsequently, an Arduino is utilized to measure the

voltage after the smoothing process.

The "/force_calculate_node" converts the

measured strains into the corresponding operating

force. Upon invocation, this node conducts

calibration to determine the parameters in the offsets

matrix B. Following calibration, it performs the

conversion based on Equation (1) and Equation (2) as

explained in Section 2.1. This conversion process

enables the calculation of the operating force from the

measured strains. The calculated operating force is

subjected to software smoothing using the moving-

average method with a window size of n = 13. This

smoothing technique helps to reduce noise and

fluctuations in the operating force data. The delay of

these procedures is designed to be less than 100 [ms],

considering the subjective evaluation of the

perception of delay time, as discussed in the study by

(Miyasato, 1995).

The "/intention_estimate_node" estimates the

user's operational intention based on the measured

operating force. In estimating the traveling direction

of translatory movement, the system calculates the

resultant force vector between the x-component and

y-component of the operating force. The system

assumes that the direction of the resultant force vector

represents the traveling direction of the cart. This

assumption allows for determining the intended

direction of movement based on the calculated force

vector. However, it is worth noting that the resistance

of the strain gauge may also change undesirably in the

direction where no force is applied due to the

Figure 6: ROS network in the steering-assist cart on

translational movement.

Table 1: Hardware components.

structural stretching. For instance, during straight

translatory movement in the positive x-direction, it is

observed that the resistance of the strain gauge on the

y-axis also changes, leading to variations in the y-

component of the operating force. Consequently,

these changes in the y-component of the operating

force can create a discrepancy between the resultant

force vector and the user's intended traveling

direction. The system incorporates a process to

determine whether the deformation is significant or

insignificant to address this issue. This judgment is

carried out by comparing the difference from a

predetermined threshold, as specified in Equation (3).

The method of determining the threshold is described

in the previous paper (Aoki, 2023).



𝐹



𝐹





𝐹



𝐹





𝐹



0



𝐹



𝐹





(3)

The "/front_command_node" and

"/rear_command_node" calculate the target steering

angle to control the active steering caster. In the case

of translatory movement, the target angle for the

Function Hardware Parts / Model No. Total

Strain Gauge FLA-5-11-3LJC 6

Strain Amp STA-12L 3

Input Controller Arduino Uno 1

Computer Raspberry Pi 4 1

Output Controller Arduino Leonardo 2

Motor Driver CMD2120P 2

Stepping Motor PKP258U20AA2 2

Electromagnetic Clutch 102-05-11 2

Operational

Interface

Compute

Active

Steering

Caster

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

692

active steering caster is set to be the same as the

estimated traveling direction. Therefore, the system

determines the estimated traveling direction and

translates it into the target steering angle, which is

then used to command the respective front or rear

active steering caster. If the target angle is too fine, it

can lead to unstable vibrations in the steering axis

caused by noise. To ensure system stability, our

approach involves determining the target steering

angle at regular intervals of 𝜋/12. We derived this

interval through a trial and error, considering the

system's response and stability. By discretizing the

target steering angle in this manner, we mitigate the

noise impact and minimize the occurrence of unstable

vibrations in the steering axis.

The "/front_control_node" and

"/rear_control_node" control the active steering

caster to align with the commanded target steering

angles. These nodes are implemented on an Arduino

microcontroller, utilizing a timer interrupt with a

cycle duration of 800 microseconds. During each

interruption, the system compares the current steering

angle with the target steering angle and determines

the appropriate rotation direction to minimize the

difference between the two angles. Subsequently,

suppose a rotation is necessary to align the current

steering angle with the target steering angle. In that

case, the system issues a command to the motor driver

to initiate a one-step rotation. The stepping motor

used in the system has a stepping angle of







[rad], and the gearing ratio between the motor and the

steering axis is 8. As a result, the rotation speed of the

steering axis is calculated to be 23.4 rpm. This

rotation speed was determined based on experimental

measurements of the steering speed during passive

steering.

4 PERFORMANCE EVALUATION

This chapter assesses the effectiveness of the steering

assist system for translational movement discussed in

the previous chapter. The steering assist system on

translational movement requires two functions. One

function generates the steering force to change the

direction of movement, while the other supports

corrective steering. We evaluate these functions using

subjective and objective evaluation indicators.

For the performance evaluation, we utilize a

prototype cart loaded with a 30 kg experimental

eccentric weight. The weight is intentionally designed

to create eccentricity in the load. This cart can switch

between passive and active steering modes. In this

experiment, the passive steering is also evaluated as a

control for the proposed system. We compare the

active and passive steering methods' operating force

and subjective evaluation scores to assess their

performance. The subjects first operated the cart with

one of the two steering methods. Then, after a one-

week interval, they operated the cart with the

remaining other steering method. We divided the 16

subjects (14 men and 2 women, aged 23±1) into two

groups, ensuring no significant differences in body

height, dominant hand, or sex. One group of subjects

started the experiments with passive steering, while

the other started with active steering. The

questionnaire items for subjective evaluation are as

follows: (A) How would you rate the operation in

terms of heaviness? (B) How would you rate the

operation in terms of intuitiveness? (C) How would

you rate the operation in terms of the cart's non-

swaying motion? (D) How would you rate the

steerability? Subjects answer each questionnaire item

using a five-score scale ranging from -2 to 2



2,1,0,1,2



4.1 Evaluation of Steering Force

In this section, we assess the operability of changing

the moving direction of the cart. Figure 7 illustrates

the experimental setup and method used in this

evaluation. At the start of the experiment, all caster

wheels are initially aligned in a same straight

direction (x-direction). After that, subjects move the

cart in a cross direction for one meter. The evaluation

indicators include the maximum operating force in

the y-direction 𝐹







and the questionnaire items

(A) ~ (D). We evaluated the differences in each

evaluation indicator using a two-way analysis of

variance (ANOVA), considering the factors of

Figure 7: Experimental setup and method employed for

evaluating steering force.

Development of Cart with Providing Constant Steerability Regardless of Loading Weight or Position: 3 rd Report on Evaluation of a

Steering Assist System on Translational Movement

693

steering method (active or passive steering) and

moving direction (left and right) in a total of four

groups. Each subject performs this experiment twice,

using each steering method and moving direction (±y)

combination.

4.1.1 Objective Evaluation

Figure 8 displays a box plot illustrating the maximum

operating force required when changing the direction

of movement. That data is measured 32 times (each

subject experimented twice) for each steering method.

The two-way ANOVA indicates that active steering

requires less operating force than passive steering

when changing the direction of movement. The result

of multiple comparisons using the Tukey method

reveal significant difference in the maximum

operating force for passive steering across different

directions. However, for active steering, there was no

significant difference between them. It is suggested

that active steering can achieve consistent steerability

regardless of the moving direction, even when being

affected by the loading weight and position. This

suggests that our system is superior to a conventional

cart system.

Figure 8: The Maximum operating force 𝐹







during

directional movement of the cart (shown in Fig. 7).

4.1.2 Subjective Evaluation

Figure 9 presents a box plot illustrating the subjective

evaluation scores for each questionnaire item. The

two-way ANOVA reveals that active steering

Figure 9: The subjective evaluation scores during

directional movement of the cart.

outperforms passive steering, as there are significant

differences in all questionnaire items. The multiple

comparisons using the Tukey method indicate that

there is no significant difference between left and

right-moving when using the same steering method.

However, in the case of different steering methods,

most questionnaire items show significant differences

at a 1% significance level. Therefore, it is suggested

that the proposed system is superior in terms of

subjective evaluation during directional movement of

the cart.

4.2 Evaluation of Corrective Steering

In this section, we assess the operability of

translational movement when the cart is affected by

eccentricity resulting from the loading position.

Figure 10 illustrates the experimental setup used to

evaluate the operability of translational movement. In

this experiment, the weight is positioned eccentrically

in the y-direction. At the experiment's beginning, all

casters' wheel orientations are aligned in a same

straight direction (x). After that, subjects move the

cart in a straight direction (+x) for a distance of four

meters. The evaluation indicators include the

maximum operating (pushing) force in the x-direction,

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

694

the standard deviation of the operating force in the y-

direction, and the questionnaire items (A) ~ (D). The

reason for incorporating the standard deviation of the

operating force in the y-direction is that the corrective

steering effect causes the operating force to fluctuate

in the y-direction in case there is an effect of the

eccentricity weight. Therefore, the standard deviation

is used as an indicator to assess this variability. Each

subject performs this experiment twice, using each

steering method.

Figure 10: Experimental setup and method employed for

evaluating corrective steering

4.2.1 Objective Evaluation

Figure 11 is a box plot showing the maximum

operating force in the x-direction and the standard

deviation in the y-direction during the straight

movement. That data is measured 32 times (each

subject experimented twice) for each steering method.

The data indicate that active steering requires less

operating force than passive steering. Furthermore,

the standard deviation of the operating force in the y-

direction is lower for active steering compared to

passive steering. It suggests the steering assist cart

requires less corrective steering than a conventional

cart.

4.2.2 Subjective Evaluation

Figure 12 shows the results of the subjective

evaluation scores for each questionnaire item. Our

analysis reveals that, except for stability, there are

significant differences between active and passive

steering in terms of subjective evaluation scores, with

a significance level of 5%. Based on these findings,

we can confirm that active steering is superior to

passive steering.

Figure 11: In case of pushing the cart in x direction with

eccentric weight (a) The maximum pushing force 𝐹





(b) The standard deviation of lateral force 𝜎





Figure 12: The subjective evaluation scores of translational

movements during pushing the cart in x-direction with

eccentric weight.

Development of Cart with Providing Constant Steerability Regardless of Loading Weight or Position: 3 rd Report on Evaluation of a

Steering Assist System on Translational Movement

695

5 CONCLUSION

In this paper, we have implemented a steering assist

system for translational movement to achieve

consistent steerability regardless of the loading

weight and position. Objective and subjective

evaluations of this system were then conducted.

Initially, we briefly introduced the proposed

active steering caster and operating interface. We

then explained the ROS nodes and network used to

realize the steering assist system, including detailed

information about each function's algorithm and

hardware specifications.

Subsequently, we evaluated the implemented

steering assist system objectively and subjectively. In

the objective evaluation, we confirmed that the

system effectively supports the operating force and

corrective steering. In the subjective evaluation, we

verified that the proposed system reduces the

perception of weightiness and enhances the

intuitiveness of the operation for users. In other words,

the provided constant steerability makes shopping

more comfortable and safer.

A limitation of this paper is that the effectiveness

was confirmed only for assisting the parallel motion

of the carts. Therefore, our future work is to

implement the steering assist system for rotational

movement. To achieve this, we will model the

relationship between the angles for diagonally

arranged active steering casters and the center of

rotation in the cart. Additionally, we will devise a

method to estimate the center of rotation based on the

measured operating force from the operational

interface. Furthermore, we will develop a control

system to ensure that the steering angles of the active

steering casters align with the desired center of

rotation and moving direction. These will enable a

steering assist system capable of omnidirectional

translations and rotation.

REFERENCES

Islam, M. M., Lam, A., Fukuda, H., Kobayashi, Y., &

Kuno, Y. (2019). A Person-Following Shopping

Supported Robot Based on Human Pose Skelton Date

and LiDAR Sensor. ICIC2019, 11645, 9–19.

Maeda, H., Fujiwara, S., Kitano, H., Yamashita, H., &

Fukunaga, H. (2003). Control of an Omni-Directional

Power-Assisted Cart. JSME International Journal

Series C Mechanical System, Machine Elements and

Manufacturing, 46(3), 931–937.

Hirata, Y., Hara, A., & Kosuge, K. (2007). Motion Control

of Passive Intelligent Walker Using Servo Brakes.

IEEE Transaction on Robotics, 23(5), 981–990.

Yun, S., Wen, W., An, Q., Hamasaki, S., Yamakawa, H.,

Tamura, Y., Yamashita, A., & Asama, H. (2017).

Investigating the Relationship between Driver’s Sense

of Agency and EEG: Mu-Rhythm Is More Suppressed

in Higher SoA Case. 2017 International Symposium on

Micro-NanoMechatronics and Human Science (MHS),

1–6.

Nakayama T., Wada M. (2023). Study on an Add-on Type

Electric Wheelchair Using Active Caster with the

Differential Mechanism. Jounal of Robotics and

Mechatronics. 2023, 35(1), 99-112

Aoki, S., Yokota, S., Matsumoto, A., Chugo, D.,

Muramatsu, S., & Hashimoto, H. (2023). Development

of Cart with Constant Steerability Regardless of

Loading Weight or Position: 2nd Report: Proposal of

the Operational Interface. 2023 IEEE International

Conference on Mechatronics (ICM), 1–6.

Aoki, S., Yokota, S., Matsumoto, A., Chugo, D.,

Muramatsu, S., & Hashimoto, H. (2022). Development

of Cart with Constant Steerability Regardless of

Loading Weight or Position: 1st Report: Proposal of an

Active Steering Caster and Its Arrangement. IECON

2022 – 48th Annual Conference of the IEEE Industrial

Electronics Society, 1–6.

Ueno, Y., Kitagawa, H., Kakihira, K., Sakakibara, T., &

Terashima, K. (2014). Development of an Innovative

Power-Assist Omni-Directional Mobile Bed

Considering Operator’s Characteristics. International

Journal of Automation Technology, 8(3), 490–499.

Seino, A., Wakabayashi, Y., Kinugawa, J., & Kosuge, K.

(2017). Control Method of Power-Assisted Cart with

One Motor, a Differential Gear, and Brakes Based on

Motion State of the Cart. 2017 IEEE/RSJ International

Conference on Intelligent Robots and System (IROS),

2829–2834.

Miyasato, T., & Kishino, F. (1995). Subjective Evaluation

of Perception of Deley Time between Visual

Information and Tacticle Information. The Institute of

Image Information and Television Engineers, 49(10),

1353–1356.

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

696