Testing Safer School Streets Before They Are Built: Agent-Based Mobility Simulation in PROBONO

Around schools, streets are more than transport corridors. They are arrival points, crossing spaces, waiting areas, social spaces, and daily zones of interaction between children, parents, pedestrians, cyclists, buses, and private vehicles. Even small changes in street layout or traffic rules can influence how people move, where vehicles slow down, and how safe the environment feels.

But testing such changes directly in the real world is not always easy. Redesigning a street, changing circulation, removing parking spaces, or introducing traffic-calming measures can be costly, disruptive, and difficult to reverse. For this reason, digital simulation offers an important opportunity: it allows planners to test possible interventions virtually before deciding what should be implemented on the ground.

Within the PROBONO H2020 project, the Technical University of Crete developed an Agent-Based Mobility modelling approach for the Brussels Living Lab, focusing on the school corridor around the ACE school. The work supports one clear user story: “As a city planner, I want to quickly simulate the impact of specific road design interventions on mobility and safety KPIs.”

The overall workflow of this approach is shown in Figure 1, linking the safety report, traffic-count data, scenario development, SUMO simulation, and KPI evaluation.

Figure 1: Workflow of the SUMO-based agent-based mobility modelling approach used to evaluate school-corridor redesign scenarios in the Brussels Living Lab.

This means that instead of relying only on static drawings or general assumptions, proposed street-design changes can be translated into a digital traffic model, tested under comparable conditions, and evaluated using measurable indicators such as vehicle speed, travel time, waiting time, stops, delay, and emissions.

Why agent-based modelling?

Urban traffic is complex because it emerges from many individual decisions. Each driver, pedestrian, cyclist, bus, or delivery vehicle reacts to the surrounding environment: a crossing pedestrian, a slower vehicle ahead, a bus stop, a narrowed road segment, a traffic rule, or a change in priority.

Agent-Based Modelling, or ABM, captures this logic by representing individual road users as “agents” inside a simulation. Instead of treating traffic as one uniform flow, the model simulates many individual movements and interactions. This is particularly useful in school areas, where local details matter: where people cross, how fast vehicles approach, whether drivers slow down near a school frontage, and how traffic-calming measures affect flow.

For the Brussels school corridor, the modelling work used Simulation of Urban Mobility (SUMO), an open-source microscopic traffic simulation tool. SUMO allows the creation of a virtual road network where different road users can move through the area and respond to changes in the street configuration. In the PROBONO study, this made it possible to compare the existing street situation with several proposed intervention scenarios.

From real-world observations to a digital baseline

The first step was to create a baseline model of the school corridor and adjacent streets. This baseline represents the current or reference situation against which all future scenarios can be compared.

To make the model more realistic, the simulation was calibrated using observed school-day traffic counts. These data included cars, bikes and motorcycles, pedestrians, and large vehicles. Pedestrian flows were also calibrated explicitly to reflect the same day and time-of-day pattern. This is important because school areas are not constant throughout the day: morning arrival, afternoon departure, and daytime periods can produce very different mobility patterns.

Figure 2 shows an example of the observed school-day traffic-count patterns used to support the calibration of the digital baseline model.

Figure 2: Example of observed school-day traffic-count patterns used to support calibration of the digital baseline model.

The model was then run for a full 24-hour simulation period. This allowed the project team to assess not just a single moment, but the daily operational behaviour of the school corridor. The baseline therefore acts as a sensor-faithful operational reference case for comparing all intervention scenarios.

The emissions assessment was also included in the modelling workflow, using Belgium 2025 passenger-fleet assumptions and HBEFA-based emission classes. This means that the model did not only look at traffic performance, but also at environmental indicators such as CO₂, CO, hydrocarbons, particulate matter, and NOx.

The scenario family: from light-touch improvements to stronger redesigns

The study tested a family of scenarios representing different levels of intervention.

Scenario A represents an enhanced 30 km/h school-zone approach. It focuses on operational and visibility improvements, such as better recognisability of the school environment, improved signage, and traffic-calming behaviour without major geometric redesign. It can be understood as a low-cost upgrade that aims to make drivers more aware and cautious.

Scenario B builds on Scenario A by adding a stronger physical intervention: a chicane. A chicane changes the alignment of the road so vehicles must slow down and navigate the corridor more carefully. This represents a medium-intensity geometric traffic-calming option.

One example of this translation from a real-world traffic-calming idea into the SUMO environment is shown in Figure 3.

Figure 3: Example of how a traffic-calming intervention, such as a chicane, was translated into the SUMO simulation environment.

Scenario C represents a shared-space concept. In practical terms, the final simulation used the best stable SUMO proxy for this concept: a 20 km/h corridor, bike mixing in the carriageway, low-priority crossings, a school-frontage crossing aligned with the detector area, and a sublane model. It should be understood as a simplified but stable representation of shared-space behaviour, rather than a literal reconstruction of pedestrians moving freely everywhere along the frontage.

Scenario D small and Scenario E represent one-way circulation options based on concept figures. Both include bus and bicycle counterflow, but they are interpreted carefully as local corridor-level proxies. They are useful for understanding how access filtering may affect the immediate school frontage, but they should not be read as full network-wide circulation studies.

This distinction is important. Some interventions affect only local behaviour, while others can displace traffic to surrounding streets. A small-area model is well suited to understanding the school corridor itself, but it cannot fully describe wider rerouting effects across the municipality.

What the results show?

The simulation results highlight a central challenge in school-street redesign: improving safety conditions often involves trade-offs.

In the baseline case, the model represents the audited reference situation for the local school corridor. When intervention scenarios are introduced, the model shows how speed, delay, stop frequency, travel time, and emissions change compared with this reference.

Scenario A produces a mild traffic-calming effect. Average vehicle speed decreases slightly, while delay and waiting time increase. This is consistent with the purpose of the intervention: to encourage more cautious driving without strongly restructuring the corridor. Interestingly, Scenario A also shows slight improvement in several emissions metrics in the final official runs, suggesting that low-cost operational measures can sometimes support calmer traffic without necessarily causing a large environmental penalty.

Scenario B creates a clearer speed-reduction effect. The chicane introduces a stronger physical constraint, making vehicles move more slowly and carefully through the school area. However, this also increases time loss, stop effects, and delay. Emissions move slightly above the baseline. This is a typical result for physical traffic-calming measures: they can improve local safety conditions, but they may also increase stop-and-go movement.

Scenario C shows the strongest calming effect among the A/B/C comparison group. It produces the lowest mean vehicle speed and the highest delay and time-loss values. It also gives the highest emissions among these three scenarios. This suggests that a stronger vulnerable-user and public-space intervention may create a safer and more people-oriented corridor, but it is also the most operationally constraining option for motorized traffic.

Scenarios D small and E show strong local reductions in delay, stops, and emissions inside the clipped modelled area. However, these results must be interpreted with caution. The apparent improvements are partly caused by reduced motor traffic inside the local study area after one-way access restrictions are introduced. In other words, the modelled corridor becomes calmer partly because fewer vehicles pass through it. This does not automatically mean that the wider network performs better, because some traffic may be rerouted outside the modelled area.

This is why the study treats D small and E as corridor-level access-filtering proxies, not as full circulation solutions.

Why the model scale matters?

The team also explored a larger-area modelling option. In principle, a larger model could better capture detours, rerouting, displaced congestion, and wider circulation effects. However, the larger-area attempts were not retained as the official study basis.

There were two main reasons:

First, the larger-area models were less stable. Simulation instability appeared through issues such as high numbers of vehicle teleports and collisions in some attempts. These are practical indicators that the model is not yet robust enough to serve as the official comparison basis.

Second, a larger-area model would dilute the policy question. The focus of this study is explicitly local: what happens around the school frontage and nearby streets? If the model area becomes too large, the indicators start to include unrelated movements, longer trips, and congestion effects outside the school corridor. This can make the results less directly useful for evaluating the school-zone intervention itself.

For this reason, the calibrated small-area model was considered the most defensible official basis for the current study. It is better aligned with the local decision question and provides a clearer interpretation of corridor-level impacts.

Supporting the PROBONO Global Digital Twin

Beyond the Brussels case study, this work contributes to the wider ambition of the PROBONO Global Digital Twin. The goal is not only to run one isolated traffic simulation, but to develop a replicable workflow that other Living Labs and urban contexts could adapt.

The methodology combines three important ingredients:

·      real-world traffic sensing,

·      expert safety assessment,

·      and scenario-based simulation.

Together, these elements allow users to move from observed mobility conditions to digital testing of possible interventions. In the future, this process can support planners, municipalities, and project partners in quickly assessing how street redesign options may affect mobility, safety-related indicators, and emissions.

The planned integration with the Global Digital Twin can make this process more accessible. Users could define a region of interest, prepare or adapt scenarios, run simulations, and compare key performance indicators. More detailed assessments may still require expert support, but the workflow provides a practical guide for bringing mobility simulation closer to decision-making.

A digital testbed for safer, more sustainable streets

The Brussels school-corridor study shows how agent-based mobility simulation can help cities explore design choices before implementation. It does not replace expert judgement, local consultation, or real-world planning processes. Instead, it strengthens them by providing a structured, evidence-based way to compare alternatives.

The key message is not that one scenario is universally “best.” Rather, the model helps reveal the trade-offs. Light-touch measures may improve awareness with limited disruption. Physical calming can reduce speeds more strongly but may increase delays and stop-and-go traffic. Shared-space concepts can support vulnerable users but may be operationally more constraining. One-way access filters can calm the immediate corridor but require wider network analysis before drawing broader conclusions.

For school environments, this kind of insight is valuable. It allows decision-makers to ask better questions, compare scenarios transparently, and understand the consequences of design choices before they are tested on real streets.

Through this work, PROBONO demonstrates how digital tools, real-world data, and urban planning expertise can come together to support safer, healthier, and more sustainable neighbourhoods.

---