A Social Force Model of the Evacuation from a Big Box Store

This paper presents an extended anisotropic social force model that incorporates realistic body shapes, physical contact forces, and behavioral factors like panic and herding to simulate a big-box store evacuation, revealing that restricting egress to main entrances while ignoring staff exits significantly increases evacuation times.

The Gist

Imagine a giant, chaotic dance floor inside a massive department store. Now, imagine that instead of music, a fire alarm is blaring, and everyone needs to get out as fast as possible. This paper builds a computer simulation to figure out exactly how that crowd moves, bumps, and panics.

The author, Gavin Buxton, created a "Social Force Model." Think of this model as a set of invisible rules that dictate how people behave when they are scared and trying to escape. Here is how the simulation works, broken down into simple concepts:

1. People Aren't Just Circles; They Are Shapes

In many old computer games, people are just simple circles. But in real life, we have shoulders, and we walk sideways to squeeze past others.

• The Analogy: Imagine people as ellipses (like flattened ovals) rather than perfect circles. This helps the computer understand that you can slide your shoulder past someone else without crashing.
• The Wheelchair Factor: The model also includes people in wheelchairs. Instead of circles, these are represented as irregular polygons (jagged shapes), like a box with wheels. This accounts for the fact that a wheelchair takes up more space and moves differently than a walking person.

2. The Invisible "Push and Pull"

The simulation uses two main types of "forces" to move people around:

• Social Forces (The "Personal Space" Bubble): This is the invisible urge to not bump into people. It's stronger if someone is in front of you than if they are behind you. The model even includes a rule that makes people naturally drift to the right to pass each other, just like how we do in real life.
• Physical Forces (The "Squeeze"): When the crowd gets too thick, people actually touch. The model calculates the "squish" (normal force) when bodies overlap and the "friction" (tangential force) when people try to slide past each other while stuck.

3. Panic is Like a Contagion

The model doesn't just make people run faster; it simulates how fear spreads.

• The Contagion: Panic spreads like a cold. If you see someone running in a panic, or hear them screaming, your own panic level goes up.
• The Herd: When panic gets high, people stop thinking for themselves. They start following the crowd blindly, even if there is a closer, empty exit right next to them. This is called "herding."
• The Freeze: Interestingly, the model notes that sometimes panic makes people freeze, but mostly it makes them run faster (up to a point).

4. The "Big Box Store" Experiment

The author tested this model in a virtual "Big Box Store" (like a giant Walmart or Target) with seven different exits:

• The Main Doors: The big front doors everyone uses.
• The Hidden Doors: Staff-only exits, emergency doors in the back, and side doors near the garden center.

The Big Discovery:
The simulation found that if people only use the two main front doors (ignoring the other five exits), the evacuation takes significantly longer.

• Why? It's like trying to pour a gallon of water out of a bottle through a tiny straw instead of the wide mouth. Even though the main doors are familiar, the crowd gets so jammed up that everyone gets stuck.
• The Solution: If people are willing to use the "staff only" or side exits, the crowd spreads out, and everyone gets out faster.

5. Groups and Speed

• Walking in Packs: The model looked at people walking in pairs or groups of three. Surprisingly, if the group moves at an average speed, they sometimes get out faster than lone walkers because they protect each other from collisions. However, if the group is forced to move at the speed of the slowest person, they slow everyone else down.
• Disabilities: The model included people with crutches and wheelchairs. While they move slower individually, their presence didn't drastically slow down the entire crowd in this specific simulation, likely because the store was big enough to avoid total gridlock. However, in tighter, more dangerous situations, their larger size could cause more bottlenecks.

The Bottom Line

The paper concludes that in an emergency, knowing about all the exits is just as important as running fast. If people ignore the "staff only" signs or side doors and only rush toward the main entrance, they create a traffic jam that makes the evacuation slower and more dangerous. The best way to get out is to spread out and use every available door, even the ones you aren't used to.

Technical Summary

Technical Summary: A Social Force Model of the Evacuation from a Big Box Store

Problem Statement

Understanding crowd dynamics during emergency evacuations is critical for optimizing strategies that balance speed and safety, potentially reducing casualties in scenarios ranging from structural fires and terrorist attacks to mass shootings and natural disasters. While experimental elucidation of panicked crowds in real emergencies is unethical, computer simulations offer a viable alternative to predict the effects of individual characteristics, capabilities, and decision-making on evacuation efficiency.

Existing Social Force Models (SFM) typically represent pedestrians as circular agents and often overlook specific physical constraints, such as the elliptical cross-section of standing individuals or the irregular geometry of wheelchair users. Furthermore, standard models may not fully capture the anisotropic nature of social interactions (where people react differently to obstacles in front versus behind them), the propagation of panic, herding behaviors, or the complex decision-making processes involved in exit selection (e.g., ignoring "staff only" exits or switching paths due to congestion). This paper addresses these gaps by developing an enhanced SFM applied to the evacuation of a large "big-box" retail store.

Methodology

Model Framework

The study employs an agent-based Social Force Model where the dynamics of the $i$-th person are governed by the equation:
$$m_i \frac{d\mathbf{v}_i}{dt} = m_i \frac{\hat{\mathbf{e}}_i v_{0,i} - \mathbf{v}_i}{t_v} + \sum_{j \neq i} \mathbf{F}_{ij} + \sum_{w} \mathbf{F}_{iw} - \mathbf{F}_{i}^{fr}$$
Here, $m_i$ is mass, $\mathbf{v}_i$ is velocity, and $t_v$ is the acceleration timescale. The term $\hat{\mathbf{e}}_i v_{0,i}$ represents the desired velocity toward the shortest path to an assigned exit.

Key Model Extensions

1. Geometric Representation:

   • Elliptical Cross-Sections: Unlike traditional circular models, pedestrians are represented as ellipses with the semi-major axis aligned laterally (shoulders). This captures the anisotropic nature of human bodies.
   • Irregular Polygons: Wheelchair users are modeled using irregular polygons (specifically two rectangles) to accurately represent their physical dimensions and spatial occupancy.
   • Orientation Dynamics: The model includes equations for angular velocity ($\omega_i$) and orientation ($\theta_i$), allowing agents to rotate due to physical collisions and wall interactions, rather than just social tendencies.

2. Force Interactions:

   • Social Forces ($\mathbf{F}_{ij}^s$): These forces capture the psychological tendency to avoid collisions. The model incorporates anisotropy (people react less to those behind them) and a rotation matrix ($R_\phi$) to enforce passing on the right. The interaction range is extrapolated over a time $t_s$ to predict potential collisions.
   • Physical/Collision Forces ($\mathbf{F}_{ij}^c$): These depend on the area of overlap ($A_{ij}$) between agents (ellipses or polygons). They include a normal force (resisting compression) and a tangential force (resisting sliding/friction).
   • Group Interactions ($\mathbf{F}_{ij}^g$): An attractive force keeps members of social groups (dyads, triads) together, defined by a distance threshold relative to the group's center of mass.
   • Wheelchair Friction ($\mathbf{F}_{i}^{fr}$): A specific frictional force limits the perpendicular motion of wheelchairs, simulating the difficulty of moving sideways.

3. Panic and Herding Dynamics:

   • Panic Propagation: The panic level $p_i$ (0 to 1) increases based on visual cues (seeing panicked neighbors), audio cues (hearing panic), and frustration (inability to move at desired speed). Panic spreads contagiously.
   • Herding: Panicked individuals have a probability of switching their target exit to match their neighbors, leading to herd behavior.
   • Rational Switching: Non-panicked individuals may also switch exits if they perceive alternative exits as less congested and visible.

4. Decision Making:

   • Agents choose exits based on a probability function dependent on distance and familiarity. Main entrances have a higher initial probability, but agents can switch paths if congestion is perceived or if they observe others moving differently.
   • Exit paths are calculated using a visibility graph (shortest path avoiding walls).

Simulation Setup

The model was applied to a simulated big-box store layout with seven potential exits (two main entrances, and five others including "staff only" or seasonal exits). Simulations varied parameters such as panic growth rates, speed increases due to panic, and the percentage of disabled individuals (crutch users and wheelchair users). The system was iterated using the Velocity Verlet algorithm.

Key Results

1. Anisotropy and Passing Behavior

In a corridor simulation, removing the rotation matrix resulted in a bimodal distribution where agents passed on either side with equal probability. Introducing the rotation matrix ($\phi = 30^\circ$) successfully enforced passing on the right, reducing head-on collisions and aligning agents into distinct lanes.

2. Panic and Herding

Simulations demonstrated that as congestion increases, panic levels rise, leading to a transition from bidirectional lanes to a dominant unidirectional herd. Panic spreads like a contagion; as agents collide and slow down, their panic increases, causing them to follow others (herding), which further exacerbates congestion and slows the system.

3. Impact of Exit Availability

A critical finding concerns the perceived availability of exits. When the simulation restricted agents to only the two main entrances (ignoring other accessible exits), the average egress time increased by approximately 7.5 seconds compared to a scenario where all seven exits were utilized.

• Restricting exits led to higher congestion at the main doors, which paradoxically increased panic and velocity but still resulted in longer total evacuation times.
• Allowing access to alternative exits distributed the crowd more effectively, reducing congestion at the primary exits and providing shorter paths for those located near secondary exits.

4. Social Groups

When group members moved at an average speed (ignoring the "slowest member" constraint), groups actually evacuated slightly faster than individuals due to reduced collisions with other agents. However, when the model enforced that groups move at the speed of their slowest member, egress times for group members increased significantly. Furthermore, the presence of slow-moving groups negatively impacted the egress times of surrounding individuals.

5. Disabled Individuals

The inclusion of crutch users (approx. 2%) and wheelchair users (approx. 2%) resulted in longer egress times for those specific groups (wheelchair users averaged ~66s vs. ~40s for able-bodied individuals). However, the presence of these individuals had a negligible effect on the egress times of able-bodied individuals in the simulated big-box store environment, likely due to the large exit widths and low overall density preventing severe trampling or crushing.

Significance and Claims

The paper claims that the developed model provides a more physically accurate representation of evacuation dynamics by incorporating:

• Elliptical and irregular geometries for agents, moving beyond simple circular approximations.
• Anisotropic social forces that dictate passing behaviors (e.g., passing on the right).
• Dynamic decision-making where agents can switch exits based on congestion and panic levels.

The primary significance lies in the finding that ignoring "staff only" or alternative exits during an evacuation significantly increases egress times. The study suggests that in big-box store evacuations, the perceived availability of all exits is a crucial factor in optimizing evacuation speed. The authors note that while the current model handles panic and herding, future work is needed to address more aggressive scenarios involving physical injury, immobility (fallen agents), and mutual aid behaviors.