Crowd dynamics are emergent processes caused by local pedestrian interactions. In this work, we describe a decentralized crowd simulation model based on deliberative-reactive agents. The model simulates fire evacuation scenarios and includes specific behaviors such as collisions, panic, bottleneck and cluster formation. The environment interactions are passive (fire alarms) or active (adaptive guidance indicators). Results show that the model is able to generate higher level dynamics through emergence. An analysis based on psychological reaction time shows that the crowd behaves like an organic whole, aligning with observations of real-world human crowds. This paper is a summary of "Predictive Agent-Based Crowd Model Design Using Decentralized Control Systems" published in IEEE Systems Journal (2022).
Emergency evacuation of buildings is a challenging crowd behavior problem [
The dynamics of a crowd must consider the heterogeneous behaviors of individuals and their
social interactions [
Current crowd modeling studies [
The three main contributions of our study [
The aim of the model is to obtain the crowd behavior during fire evacuation. The design is
bottomup, in which agents representing persons generate high-level behaviors of the crowd through
local interactions, resulting in an agent-based simulation model (ABSM). Beside modeling
individual agent dynamics, the problem of constructing an ABSM is centered around designing
the local interaction rules so that observable patterns emerge. Rules are often at least partly
known a priori in illustrative models, whereas this paper deals with a predictive ABSM: known
agents are placed in a known environment and allowed to interact with the purpose of analysing
their emergent behavior [
Thus, we propose equipping each agent with its own decision-making inference module; in our case, a controller. This approach sustains the autonomy of agents as crowd participants and allows crowd patterns to emerge organically from the local individual autonomous decision of each agent. From the environment, the agent receives inputs modeling the the presence of obstacles, e.g., walls or smoke inhalation, as well as signals from the evacuation systems, e.g., alarms and indicators. The output is the trajectory of the agent through the environment. The network of agents with controllers form a decentralized control system (DCS). Here, the DCS is non-cooperative. This means that each agent has a locally and independently defined objective, which for a person in a crowd, is to reach an exit as fast as possible.
Figure 1 shows the architecture of the agent. We choose a deliberative-reactive controller
structure [
In [
The environment has a floor plan with diferent types of rooms connected through 4 hallways and has two exits: one through the reception hall, and one situated in the diagonally opposite corner of the plan [20]. Figure 2 shows model of the floor plan. We also model the smoke difusion, which afects walking speed [24] or the exit pedestrians choose.
The protection control system is part of the environment. Its purpose is to respond to fire events with two controllers: one for fire suppression and one for evacuation guidance. In realworld buildings, this system works alongside HVAC and other internal structural automation. The aim of the fire suppression and alarm control system is to reduce damage caused by fire, while the guidance controller aids the evacuation of occupants. It is comprised of sensing, acting, and controller agents, which emulate the smoke detectors, the alarms, the sprinkler system, and the occupant monitor. In the graphic engine, the alarms become visible as red elements when active, while sprinkler activation is a burst of blue particles.
The guidance control system we design in [
The user interface of the simulation model allows setting the number of human agents, while their initial placement on the floor plan is randomized. Agent diversity is ensured by probabilitybased decisions and internal state updates, thus obtaining behaviors of people who walk faster or slower, have diferent responses to smoke inhalation, knowledge of exit positions or levels of spatial orientation within the floor plan. Some, under the stress of evacuation, make decisions to rush to the exit farthest from them.
The human agent is represented by an entity that, in the graphic engine, can display physical properties akin to real life. For an agent, the inputs are: smoke quantity in room, state (on/of) of alarms, state (on/of) and color (green/blue) of guidance indicators, wall collision-generated forces, distance to fire, and collision-generated forces from other human agents. The internal states are: inhaled smoke, agent position, walking direction, speed, life status. The agent output is comprised of the walking direction and speed to be processed for the representation of movement on the graphical interface. The ABSM is discrete and all dynamics are computed in discrete time. The world model or knowledge base of the agent contains the location of exits and the visibility graph (world map). The internal state of the agent is updated based on previous states and inputs. Smoke inhalation afects the life status and movement speed, alarms trigger the evacuation behavior, while the controller updates the movement direction.
Motion planner: the deliberative component. During regular operation, the agent moves randomly, with self-generated directions and speeds. During drills or emergencies, the agent moves between points of the visibility graph. When the guidance system is not active, the agent makes a choice for an exit and generates its evacuation route accordingly (Dijkstra). However, pedestrians do not always choose the optimal path. A probability-based accuracy parameter models this choice. The probability of misjudgment in a semi-panic situation is up to 40% for residents without disabilities and up to 50% for persons with low stamina [25].
The reactive component is formed of a position controller subsumed with a collision avoidance behavior. The position controller governs the movement in a specified direction with a specified speed. Collision avoidance: when another agent, wall, or door frame is detected on or close to the agent direct path, a new direction is computed, followed by the recovery either toward the initial target or the next node. This behavior emulates the inattentional blindness of weaving through a crowd while influenced by emergency anxiety and panic modes, i.e. making decisions on direction under pressure [26, 27, 28, 29].
Gaussian PDF ( =49.28, =10.51) Gaussian PDF ( =89.50, =17.98) exGaussian PDF (K=15.30) exGaussian PDF (K=4.14) Exit indicators enabled
Exit indicators disabled 50 75 100 125 150 175
Total evacuation time [s] (b) Without delayed reaction.
To analyze the resulting emergent crowd behavior and to validate the ABSM we first perform the overall crowd assessment during evacuation with 142 total human agents.
Fire drill and active fire. We analyze the impact of the closest exit estimation accuracy for
pedestrians (figures and table are included in [
Crowd reaction time. In psychology, reaction time (RT) is the duration between the appearance of an exogen stimulus and the occurrence of a specific response to that stimulus [ 30]. RT is usually measured and modeled as an ex-Gaussian probability density function (PDF) [31]. We introduce crowd reaction time (CRT): the duration between the activation of alarms and the ifnalized evacuation. We ran 1000 simulations for each of four scenarios: indicators enabled or disabled (25% accuracy for ineficient decisions in panic mode), with or without human agent reaction delay. Figure 3 shows the CRT distribution is an ex-Gaussian PDF, mirroring the individual RT. The crowd behaves as an organic whole: an emergent, complex systems-of-systems.
Bottleneck formation is likely scenario is when building occupants do not have information on the location of the fire (figure 4). The efect of the pedestrian agent movement speed is illustrated in figure 4) in three cases: (a) nominal base speed for agents; (b) 40% increase; and (c) 40% decrease. We observe a change in crowd cohesion, in which panicked running leads to large clusters and trampling.
Cluster formations can be observed for various crowd sizes. Figure 5 shows frames with 80, 142, and 284 pedestrian agents in the model. These clusters emerge naturally due to collisions, diferent speeds, and common goals. The guidance controller has an efect on the crowd dynamics by enabling pedestrians to easily find the nearest exit, thus ofering external knowledge to incorporate into their own reasoning process. Otherwise, pedestrians estimate the nearest exit, resulting in the formation of collision points (figure 6).
In [
(b) 40% speed decrease.
(c) 40% speed increase.
In this study we propose a decentralized deliberative-reactive agent-based crowd model, which we simulate during fire drills and evacuation. We also design a fire suppression system and a guidance controller. Results show that the model based on agent autonomy and local interactions is able to generate higher level dynamics through emergence.
This work is partly supported by the Research Council of Norway (Sponsor Code: 288164). [19] J. Wu, H. Wang, N. Li, Z. Su, Formation obstacle avoidance: A fluid-based solution, IEEE
Systems Journal 14 (2019) 1479–1490. [20] D. Grassl, Total system eficiency: condensing boiler system myth busting, Cleaver Brooks
White Papers (accessed: 2021-11-08). [21] M. Pătraşcu, M. Drăgoicea, Integrating agents and services for control and monitoring: managing emergencies in smart buildings, in: Service Orientation in Holonic and MultiAgent Manufacturing and Robotics, Springer, 2014, pp. 209–224. [22] F. Bellifemine, A. Poggi, G. Rimassa, JADE–A FIPA-compliant agent framework, in:
Proceedings of PAAM, volume 99, London, 1999, pp. 1–12. [23] jMonkeyEngine, 2021. www.jmonkeyengine.org, Accessed: 2021-11-04. [24] M. J. Hurley, D. T. Gottuk, J. R. Hall Jr, K. Harada, E. D. Kuligowski, M. Puchovsky, J. M.
Watts Jr, C. J. Wieczorek, et al., SFPE handbook of fire protection engineering, Springer, 2015. [25] J. Koo, B.-I. Kim, Y. S. Kim, Estimating the efects of mental disorientation and physical fatigue in a semi-panic evacuation, Expert Systems with Applications 41 (2014) 2379–2390. [26] A. D. Castel, M. Vendetti, K. J. Holyoak, Fire drill: Inattentional blindness and amnesia for the location of fire extinguishers, Attention, Perception, & Psychophysics 74 (2012) 1391–1396. [27] J. Lin, R. Zhu, N. Li, B. Becerik-Gerber, Do people follow the crowd in building emergency evacuation? a cross-cultural immersive virtual reality-based study, Advanced Engineering Informatics 43 (2020) 1–13. [28] L. Fu, S. Cao, W. Song, J. Fang, The influence of emergency signage on building evacuation behavior: An experimental study, Fire and materials 43 (2019) 22–33. [29] N. Shiwakoti, M. Sarvi, G. Rose, M. Burd, Consequence of turning movements in pedestrian crowds during emergency egress, Transportation research record 2234 (2011) 97–104. [30] American Psychological Association (APA) Dictionary, http://dictionary.apa.org/ reaction-time, Accessed: 2021-11-09. [31] R. Denga, Need for speed: Applying ex-gaussian modeling techniques to examine intraindividual reaction time variability in expert tetris players, in: Proceedings of the Annual Meeting of the Cognitive Science Society, volume 43, 2021, pp. 2733–2739.