Academia.eduAcademia.edu

Outline

Title
Abstract
Figures
Introduction
Proposed Method
Discussion
Future Works and Challenges
Conclusion
References

New Pedestrian Model Based on a Social Force Model with Sub-goals

Profile image of Hirohide HagaHirohide Haga

Journal of Computers

https://doi.org/10.17706/JCP.14.11.634-649
visibility

…

description

16 pages

link

1 file

Abstract

This article proposes a new pedestrian model that precisely simulates the avoidance behavior so as to realize the pedestrians' walking action. For the design of the facility, the simulation is used as a means for predicting and analyzing the movement of pedestrians. However, most of the simulators focus on the flow of pedestrian but do not consider the movement of each pedestrian. As a result, conventional simulators are useful to simulate the phenomenon of macro level movement of pedestrian but are not useful to simulate the micro-level simulation. Therefore, we aim to implement the pedestrian flow simulator that enables us to observe each pedestrian behavior by using the Multi-Agent System. Research on pedestrian model began in the 1970s, but all models cannot simulate the behavior of the pedestrian individual well. Social Force Model, which is one of the typical pedestrian models, reproduces the behavior using the equation of motion by considering the mass point pedestrian. Accordingly, this model is effective in high-density state. By contrast, there is several problems in low-density state: the avoidance speed and trajectory. Hence, we propose a new model by introducing the concept of sub-goal and new force in order to solve this problem. Furthermore, the evaluation experiment shows that proposal model improves these problems.

Figures (25)
walking speed (V), and the pedestrian's current walking speed (Vv). Here, t is a characteristic parameter of the pedestrian. A smaller value of t means that the force acting towards the destination is larger, and a smaller value of t means that the other forces acting on the pedestrian have larger effects, thus making the pedestrian more susceptible to effects from the environment.
walking speed (V), and the pedestrian's current walking speed (Vv). Here, t is a characteristic parameter of the pedestrian. A smaller value of t means that the force acting towards the destination is larger, and a smaller value of t means that the other forces acting on the pedestrian have larger effects, thus making the pedestrian more susceptible to effects from the environment.
2.2.2. Settings field of view
2.2.2. Settings field of view
Fig. 3. Adjustment of repulsive force according to _eld of view.
Fig. 3. Adjustment of repulsive force according to _eld of view.
Tatebe et al. studied how pedestrians avoid stationary obstacles, and reported that, based on thei neasurements, the distance to an obstacle at which a pedestrian starts to take evasive action is related to th ybstacle's size and direction. They concluded that the average distance was 7.34 m for an obstacle of abou he same size as a human. Yoda et al. performed tests of how pedestrians avoid people who are walking the xther way and reported that a pedestrian's speed during an avoidance maneuver remains constan rrespective of the avoidance path. They also reported that the average distance of two pedestrians passing 2ach other is 0.7 - 1.3 m. On the other hand, there have hardly been any studies of avoidance actions involvins changes of speed. Yamamoto et al. [6] measured the movement of pedestrians on intersecting paths while yeing aware of other people and reported the results of measuring the characteristics of this behavior. The
Tatebe et al. studied how pedestrians avoid stationary obstacles, and reported that, based on thei neasurements, the distance to an obstacle at which a pedestrian starts to take evasive action is related to th ybstacle's size and direction. They concluded that the average distance was 7.34 m for an obstacle of abou he same size as a human. Yoda et al. performed tests of how pedestrians avoid people who are walking the xther way and reported that a pedestrian's speed during an avoidance maneuver remains constan rrespective of the avoidance path. They also reported that the average distance of two pedestrians passing 2ach other is 0.7 - 1.3 m. On the other hand, there have hardly been any studies of avoidance actions involvins changes of speed. Yamamoto et al. [6] measured the movement of pedestrians on intersecting paths while yeing aware of other people and reported the results of measuring the characteristics of this behavior. The
Fig. 5. Avoidance by speed selection (at start deceleration). reported that when a pedestrian decelerates to avoid another person, the speed decreases sharply to 0.8 m/s at 1.2 s after the situation shown in Fig. 5, the distance between them decreases to 1.5 m as shown in Fig. 6, and then after 0.9 s the pedestrian accelerates back to the original speed. The SFM avoidance behavior was verified based on this data.
Fig. 5. Avoidance by speed selection (at start deceleration). reported that when a pedestrian decelerates to avoid another person, the speed decreases sharply to 0.8 m/s at 1.2 s after the situation shown in Fig. 5, the distance between them decreases to 1.5 m as shown in Fig. 6, and then after 0.9 s the pedestrian accelerates back to the original speed. The SFM avoidance behavior was verified based on this data.
Fig. 7. Positional relationship and external forces between obstacle and pedestrian.
Fig. 7. Positional relationship and external forces between obstacle and pedestrian.
As Fig. 8 shows, the pedestrian was unable to avoid the obstacle in case (a) and became stuck. This is because at the starting position, as shown in Fig. 7, the force F, acting on the pedestrian from the obstacle is collinear with the force F, from the pedestrian's destination, so there is no external force acting in the x direction. On the other hand, Fig. 8 also shows that it was possible to avoid the obstacle at x = 0 in case (b). However, the pedestrian’'s distance from the obstacle at the start of avoidance was just 0.38m, which is much Fig. 8. Avoidance paths.
As Fig. 8 shows, the pedestrian was unable to avoid the obstacle in case (a) and became stuck. This is because at the starting position, as shown in Fig. 7, the force F, acting on the pedestrian from the obstacle is collinear with the force F, from the pedestrian's destination, so there is no external force acting in the x direction. On the other hand, Fig. 8 also shows that it was possible to avoid the obstacle at x = 0 in case (b). However, the pedestrian’'s distance from the obstacle at the start of avoidance was just 0.38m, which is much Fig. 8. Avoidance paths.
Fig. 10. Experimental conditions.
Fig. 10. Experimental conditions.
The sub-goal settings are made using four parameters d1, d2, d3 and 0. Parameter dj is the distance betweer the obstacle and the pedestrian, and 0 is the angle between the direction of travel and the direction of the obstacle (positive in the clockwise direction). These two parameters are used to prioritize objects when there are two or more objects regarded as obstacles. Parameter dz represents the distance between the obstacle and the sub-goal. Also, considering the acceleration effect that was mentioned as one of the issues with SFM parameter d3 represents the distance from the sub-goal beyond which the pedestrian's goal is switched fror the sub-goal back to the original destination. When the obstacle is another pedestrian, this switching of goal: takes place at the end of each step regardless of the magnitude of dz, because the distance changes at eact step. These parameters are illustrated in Fig. 12.
The sub-goal settings are made using four parameters d1, d2, d3 and 0. Parameter dj is the distance betweer the obstacle and the pedestrian, and 0 is the angle between the direction of travel and the direction of the obstacle (positive in the clockwise direction). These two parameters are used to prioritize objects when there are two or more objects regarded as obstacles. Parameter dz represents the distance between the obstacle and the sub-goal. Also, considering the acceleration effect that was mentioned as one of the issues with SFM parameter d3 represents the distance from the sub-goal beyond which the pedestrian's goal is switched fror the sub-goal back to the original destination. When the obstacle is another pedestrian, this switching of goal: takes place at the end of each step regardless of the magnitude of dz, because the distance changes at eact step. These parameters are illustrated in Fig. 12.
Fig. 15. Simulator configuration.
Fig. 15. Simulator configuration.
Fig. 14. Simulation flowchart.
Fig. 14. Simulation flowchart.
Fig. 16. Simulator screenshot. Environment class to build relationships between the agents and the space. The space and agents that make up the MAS are defined as Environment and Intelligence classes. The user first defines all the constituent elements of the space in the Environment class. In the Intelligence class, which defines the action rules governing the behavior of agents, the agent behavior is determined by receiving information from the Environment class. Parts that are changed by the actions of agents are reflected in the Environment class to build relationships between the agents and the space. 4.2. Sub-goal Experiments
Fig. 16. Simulator screenshot. Environment class to build relationships between the agents and the space. The space and agents that make up the MAS are defined as Environment and Intelligence classes. The user first defines all the constituent elements of the space in the Environment class. In the Intelligence class, which defines the action rules governing the behavior of agents, the agent behavior is determined by receiving information from the Environment class. Parts that are changed by the actions of agents are reflected in the Environment class to build relationships between the agents and the space. 4.2. Sub-goal Experiments
Fig. 17. Avoidance paths. distance from obstacle when passing: 41.3 m, angle subtended by obstacle: 10 degree.
Fig. 17. Avoidance paths. distance from obstacle when passing: 41.3 m, angle subtended by obstacle: 10 degree.
4.3. Deceleration Processing Experiment The deceleration processing verification experiments were performed under the following initial
4.3. Deceleration Processing Experiment The deceleration processing verification experiments were performed under the following initial
Table 1. Distances Representing the Avoidance Characteristics of the Proposed Model ition Distance at start of avoidance Distance from obstacle when passing
Table 1. Distances Representing the Avoidance Characteristics of the Proposed Model ition Distance at start of avoidance Distance from obstacle when passing
As Fig. 22 shows, the pedestrians were able to pass each other under the starting conditions of case (a). In this case, the distance between the pedestrians while passing was 0.91 m. At 2.3 s after the start of strong deceleration, the minimum speed of 0.72 m/s is reached, after which the pedestrian accelerates back to
As Fig. 22 shows, the pedestrians were able to pass each other under the starting conditions of case (a). In this case, the distance between the pedestrians while passing was 0.91 m. At 2.3 s after the start of strong deceleration, the minimum speed of 0.72 m/s is reached, after which the pedestrian accelerates back to
Fig. 20. Pedestrian paths.
Fig. 20. Pedestrian paths.
Fig. 22. Speed variation of case (b). Fig. 21. Speed variation of case (a).
Fig. 22. Speed variation of case (b). Fig. 21. Speed variation of case (a).
normal speed after 1.14 s. On the other hand, Fig. 22 shows the speed variation when simulated under the conditions of case (b). The distance from the other pedestrian at the time of passing was 1.39 m. After 1.19 s of strong deceleration, a minimum speed of 0.8 m/s was reached, and then the pedestrian accelerated back to anormal speed after 0.93 s.
normal speed after 1.14 s. On the other hand, Fig. 22 shows the speed variation when simulated under the conditions of case (b). The distance from the other pedestrian at the time of passing was 1.39 m. After 1.19 s of strong deceleration, a minimum speed of 0.8 m/s was reached, and then the pedestrian accelerated back to anormal speed after 0.93 s.
the behavior of the model shown in Fig. 23 in order to consider the issues of the proposed model. From the above results, it can be said that we were successful in reproducing the obstacle avoidance behavior of pedestrians, which was the original purpose of introducing sub-goals. However, generic simulators must be able to cope with more complex conditions such as multiple obstacles. Here, we consider the behavior of the model shown in Fig. 23 in order to consider the issues of the proposed model.
the behavior of the model shown in Fig. 23 in order to consider the issues of the proposed model. From the above results, it can be said that we were successful in reproducing the obstacle avoidance behavior of pedestrians, which was the original purpose of introducing sub-goals. However, generic simulators must be able to cope with more complex conditions such as multiple obstacles. Here, we consider the behavior of the model shown in Fig. 23 in order to consider the issues of the proposed model.

Related papers

Simulating the Collision Avoidance Behavior of Pedestrians
Franck Feurtey

2000

We present an original algorithm to model the collision avoidance behavior of pedestrians. This method is based on a representation of trajectories in (x,y,t) space. The trajectories of the pedestrians to avoid are represented in this space and each agent selects his future movements to avoid collisions, minimizing detouring and speed variation. This algorithm was implemented in a simulation of pedestrian behavior. We tuned the parameters of the model using several test situations, and we made a comparison with real-world data. Although a strictly quantitative evaluation is difficult to make, the simulated behavior appeared realistic.

View PDFchevron_right
Multi agent simulation of pedestrian behavior in closed spatial environments
Cesare Garofalo

… and Technology for …

View PDFchevron_right
Agent-Based Simulation of Pedestrian Behaviour in Closed Spaces: A Museum Case Study
S. Tudisco

Journal of Artificial Societies and Social Simulation, 2014

View PDFchevron_right
Master Thesis: Pedestrian Simulations with the Social Force Model Anders Johansson2
Anders Johansson

2004

Abstract In order to fully understand and analyze how the infrastructure in public facilities as well as in urban regions affects the overall characteristics of pedestrian crowds within these areas, computer simulations need to be performed. An implementation of the Social Force Model is discussed, which is suitable for both small-scale pedestrian simulations as well as computationally fast large-scale pedestrian simulations.

View PDFchevron_right
CrowdEgress: A Multi-Agent Simulation Platform for Pedestrian Crowd
Peng Wang

This manual introduces a simulation tool to study complex crowd behavior in social context. The agent-based model is extended based on the well-known social force model, and it mainly describes how agents interact with each other, and also with surrounding facilities such as walls, doors and exits. The simulation platform is compatible to FDS+Evac, and the input data in FDS+Evac could be imported into our simulation platform to create single-floor compartment geometry, and a flow solver is used to generate the road map towards exits. Most importantly, we plan to integrate advanced social and psychological theory into our simulation platform, especially investigating human behavior in emergency evacuation, such as pre-evacuation behavior, exit-selection activities, social group and herding effect and so forth.

View PDFchevron_right
Modeling and simulation of pedestrian behaviour: As planning support for building design
Peter Ferschin

2016

Major challenges in the simulation of pedestrians are the realistic behaviour of agents, realistic appearance and variety, and how to conveniently define larger crowds of pedestrians. In this paper, pedestrian behaviour in buildings is analysed and structured in order to develop a behavioural model for buildings with high pedestrian flows. Observing pedestrians at a larger train station is used as a basis for the creation of a three-level behavioural model. The model follows a multi-agent-based approach and consists of a strategic, tactical, and operational level. A strategy indicates why a person visits a building and is broken down into activities on the tactical level. In this model, the strategy remains constant for an agent during its lifetime, while activities and associated areas are chosen and prioritized dynamically. This results in a coarse route through the facility. The operational level consists of path planning and obstacle avoidance. The suggested behavioural model ha...

View PDFchevron_right
Representing crowds using a multi-agent model - Development of the SimTread pedestrian simulation system
Tomonori Sano

Japan Architectural Review

SimTread was developed as a pedestrian simulation system following a straightforward, highly user-definable pedestrian and spatial model. Unlike network and mesh models, this pedestrian model is a multi-agent system in which an individual pedestrian is modeled as an agent. The spatial model is based on actual coordinates, which enables the direct representation of plans using the SimTread model. The crowd characteristics represented by this system were evaluated through test cases by measuring the flow rate, which is a primary indicator of crowd evaluation. The major results are as follows: (a) SimTread demonstrated a reasonable crowd-behavior simulation in a simple plan that was equivalent to network models. Along with the added benefits of operability, such performance is highly desirable. (b) In cases where a bottleneck accumulated a number of pedestrians sufficient to block nearby exits, SimTread accurately portrayed the resulting crowd propagation. This allowed qualitative crowd analysis.

View PDFchevron_right
Implementation of Simulation Environment for Exhaustive Analysis of Huge-Scale Pedestrian Flow
Edda Klipp

SICE Journal of Control, Measurement, and System Integration, 2013

This paper describes implementation of a pedestrian simulation environment to carry out an exhaustive analysis of huge-scale pedestrian flow. Our environment consists of the CrowdWalk pedestrian simulator and the PRACTIS simulation controller. CrowdWalk has three pedestrian movement models that are based on a one-dimensional space model. PRACTIS carries out an exhaustive and efficient analysis in a cluster environment to support effective decisionmaking, enabling pedestrians to be guided according to various criteria. The authors apply CrowdWalk and PRACTIS in a firework festival to verify their guidance planning, and examine some case studies.

View PDFchevron_right
Simulating Pedestrian Movement
Renee Puusepp

This paper discusses the results of an extended pedestrian movement project carried out in the center of Tallinn, Estonia. Having gone through several rounds of in situ studies and software prototyping, we present a synthetic method for modelling data of urban qualities as perceived by individual pedestrians. The proposed method includes capturing the walkability of city elements as a landscape of explicit values. This landscape is turned into proximity fields that in turn are fed into an agent-based model for simulating the route selection mechanism of individual pedestrians. The described model is used to demonstrate how each simulated pedestrian is influenced by the perceived qualities of urban environment. Based on our observations and computational modelling experiments, we argue that pedestrian traffic in the city can be successfully simulated with agent-based models. Furthermore, such simulation models, when finely calibrated in the field observations, allow urban planners, city authorities and decision makers to evaluate the impact of future urban design scenarios to pedestrian flows in the city.

View PDFchevron_right
An Agent-Based Approach to Pedestrian and Group Dynamics: Experimental and Real World Scenarios
Giuseppe Vizzari

The simulation of pedestrian dynamics is a consolidated area of application for agent-based models: successful case studies can be found in the literature and off-the-shelf simulators are commonly employed by decision makers and consultancy companies. These models, however, generally do not consider the explicit representation of pedestrians aggregations (groups), the related occurring relationships and their dynamics. This work is aimed at discussing the relevance and significance of this research effort with respect to the need of empirical data about the implication of the presence of groups of pedestrians in different situations (e.g. changing density, spatial configurations of the environment). The paper describes an agent-based model encapsulating in the pedestrian's behavioural specification effects representing both traditional individual motivations (i.e. tendency to stay away from other pedestrians while moving towards the goal) and a simplified account of influences related to the presence of groups in the crowd. The model is tested in a simple scenario to evaluate the implications of some modeling choices and the presence of groups in the simulated scenario. Moreover, the model is applied in a real world scenario characterized by the presence of organized groups as an instrument for crowd management. Results are discussed and compared to experimental observations and to data available in the literature.

View PDFchevron_right

References (9)

  1. Kaneda, T. (2010). Pedestrian Agent Simulation by using Artisoc. Tokyo: Kozo-Keikaku Engineering, Inc. Publishers.
  2. Helbing, D., & Molnar, P. (1995). Social force model for pedestrian dynamics. Physical Review E, 51, 4282- 4286.
  3. Helbing, D., Farkas, I. J., Molnar, P., & Vicsek, T. (2002). Simulation of pedestrian crowds in normal and evacuation situations. In M. Schreckenberg, and S. D. Shama (Eds.), Pedestrian and Evacuation Dynamics (pp. 21-58). London: Springer.
  4. Tatebe, K., Tsujimoto, M., & Shida, K. (1994). Methods for judging the beginning point of avoiding behavior and avoidance distance between a pedestrian and a standing obstacle: A study on pedestrian behavior of avoiding obstacles (II) Journal of Architecture, Planning and Environmental Engineering, Architectural Institute of Japan, no. 465, pp. 95-104.
  5. Yoda, M., & Shiota, Y. (1998). Motion characteristics and avoidance motion algorithm based on human passing behavior: Results of experiments under natural and laboratory settings [in Japanese]. Transactions of the Japan Society of Mechanical Engineers, 64(619), pp. 959-965.
  6. Yamamoto, T., Shibata, S., & Jindai, M. (2006). An analysis of human avoidance behaviour of a collision by deceleration in crossing each other. Japan Ergonomics Society, 42(6), pp. 402-407.
  7. Yanagisawa, Y., Yamada, T., Hirata, K., & Satoh, T. (2006). Pedestrian dynamics with decision of Subgoals based on eye-gaze. IPSJ Journal, 47(7), 2160-2167.
  8. Tohru Mizuno received his B.Eng and M.Eng in information and computer science from Doshisha University in 2012 and 2015 respectively. He also received the Diplome d'Ingenieur (French engineering degree) from Ecole Centrale du Lille, Lille, France in 2014. After finished his master's study, he joined to Hitachi, Ltd.. Currently, he is an engineer of System Control Engineering Division of Hitachi, Ltd.
  9. Hirohide Haga received his B.Eng and M.Eng in electronics from Doshisha University and Ph.D. in computer science from Kyoto University in 1978, 1980, and 1996 respectively. After finishing his master's study in 1980, he joined in Hitachi, Ltd., where he was a research staff member from 1980 to 1994. In 1994, he moved to Doshisha University as a faculty member. Currently, he is a full professor of computer and information science at Doshisha University, Kyoto, Japan. Prof. Haga is a member of IEEE-CS, ACM, BCS (British Computer Society), and IEICE-J (Institute of Electronics, Information and Computer Engineers of Japan). He held several visiting positions (visiting scholar, visiting professor, and invited professor) at Imperial College of Science and Technology, University of London (UK), University of Oulu (Finland), Cambridge University (UK), E cole Centrale de Lille (France), and CentraleSupe lec (France). His research interests include software engineering, multi-agent simulation, block- chain application, and digital art. He was honored as a Chartered IT Professional (CITP) from BCS.

Related papers

An agent-based simulation model for pedestrian unidirectional movement
Wallace Rodrigues de Santana, G. Kobayashi

2009

Pedestrian movement behavior can be analyzed in different ways. One of the most common approaches for this subject is the cellular automata. However, the pedestrian behavior is usually modeled by a reactive-response framework. Also, the pedestrians are treated as just numbers or elements in a grid. This paper proposes a basic framework for a multi-agent pedestrian movement model. Applying the same rules found in a cellular-automata-reactive model, this framework allows a new look for this kind of phenomenon, and it paves the ground for a more cognitive agent model and a physical environment with distinct features as well.

View PDFchevron_right
Multiple Destinations Pedestrian Model using an Improved Social Force Model
Ryo Nishida

2018

Social Force Model is extensively used to represent pedestrian movements. This model comprises several force terms. One of the terms describes the force that is required to move to the destination. The manner to select a destination in such a model is not specified and this model is generally used for one destination. In the real world, however, the desired destination of a pedestrian is observed to change over time. Therefore, in this research, we formulate a constantly changing destination selection process and model pedestrian movement when there are multiple destinations by improving Social Force model.

View PDFchevron_right
A new autonomous agent approach for the simulation of pedestrians in urban environments
ALFONSO BRAZALEZ

Integrated Computer-Aided Engineering, 2009

In this paper, a new approach to autonomous agent navigation in real-time applications is presented. Effective driving simulators demand a realistic simulation of both the driven vehicle and the surrounding environment. For example, a variety of pedestrians and autonomous vehicles improve the realism of the simulation. Many studies of virtual environments for autonomous navigation are based on complex spatial subdivisions. Our approach takes advantage of the topological structure of cities, where pedestrians usually walk along straight pavements, to develop a new method for the simulation of autonomous pedestrians. The "smart obstacle" concept is introduced and a method for realistic corner turning is also explained. Our approach has been applied to simulate the motion of pedestrians on train station platforms, the results of which are presented and discussed.

View PDFchevron_right
Agent-based simulation of pedestrian behaviour
Matteo Ignaccolo

cesaregarofalo.it

In transport engineering, pedestrian models can be roughly separated in analytical models and micro-simulations. The first ones represents pedestrian flows through mathematical formulas providing average pedestrian flow along a street, while the second ones ...

View PDFchevron_right
Incorporating Prediction Factor into the Investigation Capability in the Social Force Model: Application on Avoiding Grouped Pedestrians
osama alia

Applied Mathematics & Information Sciences, 2013

Incorporating decision making capability into a microscopic crowd dynamic model is an essential factor for introducing a well-representative model for the pedestrian flow. In normal situations, this capability was presented by providing the simulated independent pedestrians in the Social Force Model an investigation ability of their walkways. However, predicting semi-blocked situations by the simulated intelligent pedestrians has not been treated in their investigation. In this paper, the authors describe the investigation capability of independent pedestrians in normal situations. In addition, the prediction aspect is introduced for the simulated independent pedestrians as an extension for the investigation capability to let the model appear more representative of what actually happens in reality. Simulations are performed to validate the work qualitatively by tracing the behavior of the simulated pedestrians and studying the impact of this behavior on their interactions with the grouped pedestrians. Finally, a comparison between the extended model and the original model with regards to the quantitative measurement (the efficiency of motion) is made.

View PDFchevron_right
580 California St., Suite 400
San Francisco, CA, 94104
© 2025 Academia. All rights reserved