From Mesh to Masses - On Simulations, Digital Twins, and AI
Our CEO, Roland gave a talk titled “From Mesh to Masses: On Simulations, Digital Twins, and AI” during a knowledge session hosted by de Ruimteschepper, in collaboration with AI-InfraSolutions and Kavel 10.
The event took place in a truly unique setting — a hangar at Groningen Airport Eelde — where participants could explore the practical applications of technology in aviation, vehicles, sensors, and measurement systems during the breaks. A perfect backdrop for discussions at the intersection of the virtual and real world.
He shared why simulating human crowds is not just fascinating, but essential. We explored how simulation helps optimize crowd flow at large-scale events like the Tour de France and Vuelta, how it supported social distancing measures during the 2021 Dutch elections, and how it’s used to test evacuation procedures in complex environments such as airports or during military training. These are clear examples of dual-use technology — valuable in both civilian and defense contexts.
He explained how we build these simulations, starting from the ground up. Our approach is based on a five-level behavior model, from low-level pedestrian movement up to high-level planning and decision-making. We used Kadaster and photogrammetry data to extract walkable areas. These are then populated with agents whose actions are defined by activity routes and flow groups, navigating corridors based on motion-capture-validated behavior models.
New developments that were presented include the integration of BIM data into simulations and a new plugin that extracts walkable surfaces from BIM models. We’re also incorporating generative AI to produce realistic behavior patterns and enhancing visual fidelity using real-time sensor data and photogrammetric input, which we’ve tested during events like Kingsday, the National Remembrance Day, and soon SAIL Amsterdam, through the IGV Fieldlab SAIL 2025 project.
On the streaming side, we now render simulation output directly into GIS layers — either as density maps or animated 3D characters — using formats like MWS, WFS, and even USD, taking inspiration from the animation and film industry for better interoperability.
The talk was ended with a reflection on explainability. While our AI models are well-documented and validated, their application in complex simulations often leads to emergent behavior, and such a complex system may be harder to explain. Transparency and validation through data remain crucial.
We also raised the question of how we can take digital twins to the next level of maturity. One of the biggest technical challenges is integrating all components — simulation, visualization, data streams — in a performant way.
That’s why we developed SimCrowds, a user-friendly tool for setting up and running simulations. It’s helping us and our partners gain insight into complex environments and dynamic scenarios.
#crowd #simulation #digitaltwin #AI #SimCrowds #uCrowds
Read, like and share this post on our LinkedIn!