Mobile Robot Navigation Based on Pedestrian Flow Model

Considering Human Unsteady Dynamic Behavior

Ryusei Shigemoto

and Ryosuke Tasaki

Department of Mechanical Engineering, Aoyama Gakuin University, Sagamihara, Kanagawa, Japan

Keywords: Robot Navigation, Social Interaction, Dynamic Environment, Pedestrian Flow Model.

Abstract: Achieving robot navigation that satisfies the requirements of safety and efficiency in a dynamic environment

crowded with people is a challenging task because of the need to implement social aspects of robot behavior.

In this study, a robot navigation method based on an unsteady dynamic pedestrian flow model is proposed,

taking into account the unsteady dynamic nature of pedestrian flow, which has not been taken into account in

conventional algorithms. We propose a method that enables continuous following of unsteady pedestrian flow,

which allow the robot to approach the destination safely and efficiently. The social compatibility of the

proposed navigation system, consisting of safety and efficiency, is evaluated through several simulations and

actual experiments.

1 INTRODUCTION

Navigation in densely populated environments is the

most important research topic from the viewpoint of

satisfying all the requirements of safety and

efficiency during movement since robots interact

closely with people. Trautman et al. stated that it is

important to consider human-robot interaction in

order to engage closely with people (Trautman et al.,

2013). One strategy for close interaction between

humans and robots is a navigation strategy that takes

social interaction into account (Rios-Martinez et al.,

2015). Pedestrian flow observed in crowded

environment is often formulated depending on human

social interaction and connection as shown in Figure

1 (Helbing et al., 2001; Hoogendoorn et al., 2004).

Robotic movement along the pedestrian flow reduces

pedestrian deceleration, avoidance, and crossing

behavior. Respecting the crowd flow therefore

enables safe and efficient crowd navigation.

The following three studies have been conducted

on navigation methods using pedestrian flows. Du et

al. developed a group surfing method that generates

actions that mimic pedestrian flow behavior (Du et

al., 2019). Yao et al. proposed a method for

incorporating a population-aware Social Force Model

into GAN (Yao et al., 2019). Kumahara et al.

https://orcid.org/0000-0002-4360-554X

https://orcid.org/0000-0002-3619-4498

proposed a potential method that integrates the

Lennard-Jones potential and wrapped normal

distribution (Kumahara et al., 2014). Pedestrian flow

following and merging control using the above

methods enable crowd navigation in congested

environments. However, the navigation systems

proposed in previous studies are not considerd

unsteady pedestrian flows, which may lead to failure

of robot navigation. A method for following unsteady

pedestrian flows to increase the success rate of

navigation is thus proposed in this study. The safety

and efficiency of the proposed navigation system are

then evaluated and demonstrated through several

simulations and experiments on actual equipment.

2 ROBOT NAVIGATION SYSTEM

The system configuration of experimental robot

navigation is shown in Figure 2. In this study, Terapio

(Tasaki et al. 2015), an omni-directional mobile

medical examination support robot with advanced

mobility capabilities in a dynamic crowded

environment, is used because there are no restrictions

on the direction of movement. The robot is controlled

by speed commands generated by a control PC based

on dynamic pedestrian flow information and robot

Shigemoto, R. and Tasaki, R.

Mobile Robot Navigation Based on Pedestrian Flow Model Considering Human Unsteady Dynamic Behavior.

DOI: 10.5220/0012211900003543

In Proceedings of the 20th International Conference on Informatics in Control, Automation and Robotics (ICINCO 2023) - Volume 1, pages 281-284

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

281

Figure 1: Pedestrian flows in dense human crowds.

information. At this time, the robot is controlled in the

x-axis direction of the world coordinate system and

the x

-axis direction of the robot coordinate system.

The dynamic pedestrian flow information is

consisted of pedestrian flow and pedestrian size

information. Since the use of surveillance cameras

has rapidly increased in recent years in the city,

pedestrian flow is measured using an RGB camera

(LOGICOOL STREAMCAM, Logitech CO., Ltd.)

with an overhead viewpoint similar to a surveillance

camera. Pedestrian size information is measured

using two LiDAR (UTM-30LX, Hokuyo Automatic

CO., Ltd.) mounted on the front and rear of the robot.

Thus, the pedestrian flow information is obtained by

detecting and tracking the pedestrian's head from

various color information, and pedestrian size

information is obtained by measuring the pedestrian's

stride length. The details of the algorithms for

obtaining each piece of information will be described

in another paper in the future.

3 UNSTEADY PEDESTRIAN

FLOW

A method for following unsteady pedestrian flows is

described in this section. Unsteady pedestrian flow is

defined as a flow whose velocity changes over time.

Moreover, pedestrian flow is defined as social group

of neighboring pedestrians sharing similar motion

patterns in this study. Since the pedestrian flows

observed in real-world dynamic congested

environments are unsteady, a potential function

combining the LJ potential (LJ potential) and

Wrapped Normal Distribution (WND) for steady

flows generates unnecessary following behavior.

Therefore, a potential function considered unsteady

pedestrian flow is developed as follows.

𝑈





𝑟



𝛾𝜀



𝛽



𝜎



𝑟𝑟



𝑟









𝛼



𝜎



𝑟𝑟



𝑟













(1)

Figure 2: Schematic of a system construction.

𝑓



𝑣







2𝜋𝜎





exp

𝑣



𝜇







2𝜎











(2)

𝛾𝑓



𝑣



𝑓



𝜇





(3)

𝑓





𝜃







2𝜋𝜎





exp





𝜃

𝜋

180



2𝜋𝑖

2𝜎











(4)

𝛼

𝑓





𝜃



𝑓







,𝛽



1𝑤





𝛼𝑤



(5)

where r [m] is the distance between the robot and the

pedestrian; ε

, p, and q are parameters that adjust the

magnitude of the proposed potential function, the

repulsion term, and the attraction term; σ

is a

parameter that adjusts the approach distance to the

pedestrian to be tracked; r

[m] is the robot radius;

[m] is the pedestrian radius; σ

is the standard

deviation of the normal distribution; μ

flow

[m/s] is the

pedestrian flow velocity to be tracked. v

[m/s] is the

pedestrian velocity; θ [°] is the angular difference

between the robot's destination direction and the

pedestrian's direction of travel; σ

WND

is a parameter

that determines how far away from the robot

destination direction and the pedestrian's direction of

travel the robot will follow or not follow; w

WND

is a

parameter that adjusts the magnitude of the repulsion

term in the proposed potential function.

(1) is a potential function that combines the LJ

potential and WND, weighted by (3). The outline of

the potential graph generated by (1) is shown in

Figure 3. By weighting (1) with (5), the robot follows

pedestrians moving in a direction similar to the

destination direction and avoids others. Moreover, by

weighting (1) with (3), it is possible to reduce the

potential effect generated by unsteady pedestrians

moving at a velocity different from that of the

pedestrian flow. Therefore, the potential function that

is considered unsteady pedestrian flows generates a

continuous following motion for the unsteady

pedestrian flow.

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

282

(a) Conventional system.

(b) Proposed system.

Figure 5: Results of simulation experiments with the

proposed system and comparison system.

Pushed back

Follow without overtaking

Merging

Follow and overtake

4 CROWD ENVIRONMENT

SIMULATION

The safety and efficiency of the proposed system

were evaluated with several simulations in a

congested environment with unsteady pedestrian

flow. Simulations are performed in the environment

shown in Figure 4. The robot moving velocity is 1.3

m/s, and the pedestrian velocity is 1.0 m/s. However,

if the robot x-coordinate exceeds 6 m, the velocity of

one pedestrian in front of the robot is set to 0.5 m/s.

Figure 5 shows the simulation results of the

conventional system (Kumahara et al., 2014) and the

proposed system. It is important to note that the

proposed system is implemented with a pedestrian

flow merging algorithm to increase the navigation

success rate, but the details of the algorithm will be

discussed in a future paper. The conventional system

is not implemented with the merging algorithm and is

not considered unsteady pedestrian flows. Therefore,

the robot fails to merge with the pedestrian flow

heading in the direction of the destination and is

pushed back as shown Figure 5(a). Furthermore, it

does not overtake and follows pedestrians moving

slower than the pedestrian flow velocity. These

movements were unnecessary movements. The

proposed system is implemented with a merging

algorithm and considered unsteady pedestrian flows.

Therefore, it is confirmed that the robot is heading

toward the confluence point where it can safely and

efficiently travel to the most destinations as shown

Figure 5(b). It also overtook pedestrians moving

slower than the pedestrian flow velocity and

continuously followed the unsteady pedestrian flow.

The above results show that the proposed system

takes less time and travels shorter distances to the

destination than the conventional system.

Furthermore, the generated robot trajectory is smooth.

Such robot behavior is considered safe and

efficient. Therefore, it can be evaluated that the

proposed navigation system is safer and more

efficient than the conventional navigation system by

implementing sociality that continuously follows

unsteady pedestrian flows.

5 NAVIGATION EXPERIMENT IN

UNSTEADY PEDESTRIAN

FLOW

An experiment based on dynamic pedestrian flow

information is conducted using an omnidirectional

mobile robot in an environment with dynamic

pedestrian flow. In the experimental environment,

there are four pedestrians moving at 0.5 m/s in the

direction of the destination and one pedestrian

moving at 0.5 m/s in the opposite direction.

Figure 4: A simulation environment.

16 m

6 m

Destination

Pedestr ian flow

Pedestr ian

Robot

Robot velocity

Figure 3: Potential graph.

Distance between target person and robot r [m]

unsteady

(r) [-]

Mobile Robot Navigation Based on Pedestrian Flow Model Considering Human Unsteady Dynamic Behavior

283

Figure 6: Results of experimental navigation using dynamic

pedestrian flows.

Time = 0 s

Destination

Robot position

(a) Star t navigation

Time = 3 s

Merging point

Pedestrian velocity

Pedestrian flow

(b) M erging the pedestrian flow

Time = 10 s

Pedestrian detect ed

position

Pedestrian position

Time = 15 s

(d) Reaching the destination

The actual navigation results are shown in Figure

6. The left side is a rendered image of the pedestrian

flow information, and the right side is a robot motion

image viewed from the side. In the rendered image,

the robot position is drawn as a coordinate system, the

pedestrian head detected by the overhead view

camera is depicted as a white circle, the detection area

is shown as a green rectangle, the pedestrian position

as a red circle, the pedestrian velocity is indicated as

a blue arrow, the pedestrian size is marked as a green

circle, the pedestrian flow is illustrated as a large

green rectangle, the merge point is indicated as an

orange circle, and the destination point is marked as a

white star. The robot successfully merges into the

optimal pedestrian flow using the proposed system as

shown in Figure 7(b). It also overtakes pedestrians

moving slower than the pedestrian flow velocity and

continuously follows the unsteady pedestrian flow as

shown Figure 7(c). Similar to the simulation, it can be

verified that the robot action satisfies safety and

efficiency without interfering with the pedestrian

action and without taking unnecessary actions, even

when the pedestrian flow is unsteady. The safety and

efficiency of the proposed system have been

evaluated in terms of pedestrian and robot trajectories

in the current stage. In the future, we will evaluate the

proposed system from a psychological point of view

by asking subjects to complete questionnaires on

evaluation items such as safety and naturalness of the

robot navigation.

6 CONCLUSIONS

A navigation system based on an unsteady dynamic

pedestrian flow model is proposed to achieve crowd

navigation that satisfies the requirements of safety

and efficiency in a densely populated environment.

Continuous following behavior for unsteady

pedestrian flow is achieved by using a normal

distribution to reduce the potential effect generated by

pedestrians moving at a velocity different from the

pedestrian flow velocity. Social robot navigation in

congested environments with multiple unsteady

pedestrian flows can be realized by considering

pedestrian flow is dynamic and unsteady.

At this stage, experiments in specific scenes have

only been able to conduct. In the future, we will

conduct experiments assuming various scenes, such

as people staying in the flow and merging into the

flow. The usefulness of the proposed system in a

complex environment will then be verified.

REFERENCES

Trautman, P., Ma, J., Murray, R. M., Krause, A. (2013).

Robot navigation in dense human crowds: the case for

cooperation. In 2013 IEEE international conference on

robotics and automation, pages 2153-2160.

Rios-Martinez, J., Spalanzani, A., Laugier, C. (2015). From

proxemics theory to socially-aware navigation: A

survey. International Journal of Social Robotics, 7:137-

153.

Helbing, D., Molnár, P., Farkas, I. J., Bolay, K. (2001).

Self-organizing pedestrian movement. Environment

and planning B: planning and design, 28(3):361-383.

Hoogendoorn, S. P., Daamen, W. (2004). Self-organization

in walker experiments. Traffic and Granular Flow,

3:121-132.

Du, Y., Hetherington, N. J., Oon, C. L., Chan, W. P.,

Quintero, C. P., Croft, E., Van der Loos, H. M. (2019).

Group surfing: A pedestrian-based approach to

sidewalk robot navigation. In 2019 international

conference on robotics and automation (ICRA), pages

6518-6524.

Yao, X., Zhang, J., Oh, J. (2019). Following social groups:

Socially compliant autonomous navigation in dense

crowds. arXiv preprint arXiv:1911.12063.

Kumahara, W., Masuyama, G., Tamura, Y., Yamashita, A.,

Asama H. (2014). Navigation system for mobile robot

based on pedestrian flow under dynamic environment.

Transactions of the Society of Instrument and Control

Engineers, 50:58-67.

Tasaki, R., Kitazaki, M., Miura, J., Terashima, K. (2015,

May). Prototype design of medical round supporting

robot “Terapio”. In 2015 IEEE International

Conference on Robotics and Automation (ICRA),

pages 829-834.

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

284