On the night of the 2022 Halloween crowd crush in Seoul, 81,573 people left the train at Itaewon station, more than two and a half times the number recorded on the same night a year earlier. The fatal crush happened in a sloping alley about 45 metres long and roughly 4 metres wide, narrowing to around 3.2 metres, where people were trying to move in both directions at once.
Why developers and operators now ask for crowd simulation
Across the Gulf, the defining projects of the next decade are built around peak demand: Expo 2030 in Riyadh, fifteen stadiums across five host cities for the 2034 FIFA World Cup, new metro and rail lines, airport expansions, and hospitality serving millions of pilgrims in Makkah and Madinah. For these clients, how people move at the busiest minute is not a technical detail. It is the product.
Protecting people, the asset and the brand. One crowd incident can close a venue and define its reputation for years. Being able to show that the peak was tested is now part of responsible ownership.
Approvals. Performance-based fire solutions need a quantified evacuation case. Authorities increasingly expect that case to be modelled and evidenced rather than argued.
Operational readiness. Operators now run crowds with live data. During Hajj, Saudi platforms such as Baseer and Sawaher use computer vision, thermal imaging and AI analytics to track density, spot congestion and anticipate surges in real time. A design-stage model gives those control rooms their starting point: the expected densities, the trigger levels and the fallback routes.
Security and resilience. Counter-terrorism law is moving the same way. The UK’s Martyn’s Law received Royal Assent in April 2025, with active regulation expected from spring 2027. It requires qualifying premises above 200 and 800 capacity to plan for evacuation, invacuation, lockdown and communication. Every one of those is a crowd movement question, and global venue operators tend to carry such expectations into the venues they run elsewhere.
Space and cost. Evidence-based widths can release floor area where prescriptive rules are conservative, and prevent expensive late enlargement where they are not.
Why lead consultants benefit from bringing it in early
For a lead consultant or multidisciplinary design firm, crowd simulation is less about producing a report and more about controlling design risk.
Dimensions, not debates. Modelling turns circulation into widths, queue lengths and stair counts the architect can use while changing them still costs nothing.
One picture for every specialist. Fire safety, vertical transportation, security, wayfinding, landscape and operations can all test their proposals against the same model of how people move, which cuts coordination loops.
Fewer late surprises. Operator reviews and authority comments rarely arrive at concept stage. A model that has already tested the peak answers many of them before they are asked.
Stronger bids. A clear crowd strategy in a technical proposal shows a client that the design team understands how the building will be used, not only how it will look.
What crowd simulation is, in brief
Crowd simulation, also known as pedestrian or crowd modelling, represents every person as an individual agent with a walking speed, body size, destination and route choice. The software moves thousands of agents through the building geometry second by second and records where density builds, how long people queue, how fast each door, stair and escalator discharges, and how long each journey takes.
It answers two distinct questions. Emergency evacuation modelling asks whether people can reach safety before conditions become untenable, and forms part of the fire engineering case. Normal-operation modelling asks whether people can move safely and comfortably on the busiest day: ingress, post-event egress, transfers and queues at security or ticketing. The same model can test both, but the scenarios, assumptions and acceptance criteria must be kept separate.
The standards that frame a crowd study
There is no single crowd simulation code. A sound study draws on several layers of guidance, chosen for the building type.
Prescriptive egress: NFPA 101, the UAE Fire and Life Safety Code of Practice and the Saudi Building Code (SBC 801) set occupant loads, exit capacities and travel distances. Simulation is needed when a design goes beyond them or adopts a performance-based solution.
Performance-based evacuation: PD 7974-6:2019 and SFPE guidance on human behaviour describe how the Required Safe Escape Time (RSET) is built from detection, warning, pre-movement and movement time. Simulation provides the movement component; the Available Safe Escape Time (ASET) normally comes from fire and smoke CFD.
Transit: NFPA 130 requires a station platform to be cleared of its occupant load in four minutes or less, with a point of safety reached within six minutes.
Stadiums and events: the Guide to Safety at Sports Grounds (Green Guide, sixth edition) ties exit capacity to an emergency egress time of up to eight minutes depending on fire risk; FIFA stadium guidelines, the Purple Guide and the UK HSE’s HSG154 cover tournament and event planning.
Comfort: John Fruin’s Levels of Service A to F remain the common language for how crowded a walkway, stair or queue feels, and many station and venue guidelines set a minimum level for design peaks.
Model credibility: ISO 20414:2020 sets a verification and validation protocol for evacuation models, and reputable tools publish results against the IMO MSC.1/Circ.1533, RiMEA and NIST Technical Note 1822 test cases.
What recent research tells design teams
Dense crowds stop behaving like individuals
In a study published in Nature in February 2025, researchers from ENS Lyon and the University of Navarra filmed the crowd at the opening of the San Fermín festival in Pamplona across four editions. Density rose from about two people per square metre an hour before the start to around six during the event, peaking at nine. Beyond a critical density, pockets of several hundred people began swaying together in a regular cycle of roughly 18 seconds, with no push or external cue. The same motion appeared in footage from the 2010 Love Parade in Duisburg shortly before the disaster.
The practical lesson is twofold. First, at these densities a crowd behaves more like a continuous material than a set of people making choices, and conventional agent models, built on individual collision rules, were never designed to reproduce it. Second, the aim of design-stage simulation is therefore prevention: to demonstrate that a layout keeps crowds well below that regime at its worst credible peak.
Wider entry queues are not automatically safer
Controlled experiments at Forschungszentrum Jülich studied crowds in front of an entrance gate, varying the width of the barrier corridor leading to it and how motivated people were to get in. Density in front of the gate rose as the corridor became wider and as motivation increased. The narrowest corridor also felt fairest and most comfortable to participants. For stadium turnstiles, security screening and concert entries, the geometry of the queueing zone matters as much as its size.
Counterflow is a primary trigger
A 2024 simulation-based reconstruction of the Itaewon crush found average densities of about 7.6 people per square metre in the critical area, peaking near 10, and identified the collision of opposing flows as a key cause alongside the sheer number of people. When the model applied a management strategy that separated the two directions, capacity improved and the risk indicators fell. Any crowd study for a station, concourse or event route should therefore include a reverse-flow or counterflow scenario, not only a one-directional exit case.
The danger thresholds are well understood
Crowd scientists commonly place the onset of crush risk at around four to five people per square metre in a moving crowd, with danger rising sharply at six. Comfort-based design targets sit far below these figures and are normally expressed through levels of service. A good study reports both: comfort for everyday operation and a clear margin from dangerous density for every credible surge.
Where crowd problems usually begin
1. Ingress is under-designed. Security screening and ticket checks set the real capacity of many venues. Undersized screening pushes queues into drop-off zones, roads and unshaded forecourts, a serious concern in Gulf summers.
2. Vertical circulation throttles the flow. Stairs and escalators discharge far fewer people per metre than level routes, and flows merge at landings. Escalators are often excluded from emergency capacity, yet everyday operation relies on them.
3. Flows cross. Arriving and departing streams, or toilet and food queues extending into main routes, create the density hot spots that simulation exposes.
4. Everyone moves at once. Average hourly flows hide the synchronised surge: the final whistle, a train arrival, a prayer call or a flight bank.
5. The layout changes after modelling. Kiosks, security posts, seating, planters and event overlays quietly reduce the widths the model relied on.
6. Behaviour is oversimplified. Zero pre-movement, uniform speeds and no allowance for families, groups, people with reduced mobility or first-time visitors all flatter the results.
7. Animation replaces analysis. A smooth video without a scenario matrix, sensitivity tests and acceptance criteria proves very little.
A practical workflow for design teams
Concept: agree peak scenarios, occupant loads and arrival profiles with the client and operator; size gates, stairs and concourses with first-order capacity and level-of-service checks.
Schematic design: model the critical areas and test a scenario matrix covering normal peak, emergency evacuation, a blocked exit, an escalator out of service, counterflow and any event-specific surge; return results to the architect as dimensions.
Detailed design: link crowd results to the fire strategy and smoke modelling (RSET against ASET), vertical transportation, wayfinding and the security layout, and re-run after late changes.
Operations: convert the model into a crowd management plan with steward positions, signage, ingress timing, holding areas, invacuation and lockdown routes, and density thresholds for live monitoring.
Useful deliverables are practical: maximum and cumulative density maps, level-of-service maps, flow rates through each door and stair, queue lengths and waiting times, evacuation and journey times, and a register of every assumption.
Frequently asked questions
Is crowd simulation required by code?
Not as a blanket rule. It becomes necessary when a design uses performance-based egress, goes beyond prescriptive limits, or when an authority, operator or venue guideline asks for evidence of crowd safety and comfort.
How is it different from egress calculations?
Hand calculations estimate capacity from flow rates per metre of width. Simulation models individuals, so it reveals queuing, merging, counterflow and local density that hand calculations cannot show.
Can simulation predict a crowd crush?
Not reliably at extreme densities, where recent research shows crowds behave collectively. Its value lies in proving that a design keeps densities well below that range, and in setting thresholds for live monitoring.
When should crowd modelling start?
At concept stage, when stairs, gates and concourses can still change at little cost.
At Xeluxe, crowd simulation sits alongside our fire engineering and CFD work, so escape routes, smoke behaviour and everyday circulation are tested against the same design. If you are shaping a venue, station, terminal or large mixed-use scheme and want to see its peak minute before it happens, we would be glad to talk.



