 Why pedestrians do not constantly collide with other pedestrians? This question is so interesting that has been keeping us busy for the last eight years. It has to do with the so-called crowd dynamics. In crowd dynamics we study how pedestrians walk. We wonder how does this very special fluid made of human beings move? Why do we study crowd dynamics? Clearly a good understanding of the dynamics of human crowds can help in improving the design of urban infrastructures such as the airport, train or metro stations. It can also make our daily commute more efficient and safer and our leisure time more enjoyable for example by helping optimizing the flow of crowds in museums. Reliable crowd models can help automatizing crowd management in a way that it is less intrusive and more efficient. Understanding the dynamics of crowds is extremely challenging. Is it possible to model the dynamics of humans? Are we predictable and do we behave in a reproducible way at least from a statistical point of view? And how many types of different dynamics are there in different settings? These are all fascinating questions of deep fundamental interest at the crossroads between statistical physics, active matter, mathematics and psychology. As for any physical system to start understanding crowd dynamics we first need to measure it. This means acquiring trajectory information from every pedestrian for instance in a public location or tracking pedestrians. Now crowd dynamics is highly variable. It contains many different behavior. And while individual trajectory mostly reflect wheel there is highly precious information in large trajectory ensemble. To capture it we need high quality tracking data in space and time that is anonymous and that comes in sufficiently large volumes to be able to analyze statistical features. This means collecting hundreds of thousands of trajectories. We install grids of special overhead sensors that measure depth that is the distance between each point in the scene and the camera plane. Here you can observe the depth field that we acquired within Eindhoven train station. Specifically what you see here is the dense flow of the morning commuters. In the picture grey scales encode for depth. This means that heads are the darkest point whereas the floor is white. Thanks to homemade computer vision algorithm we can use this data to track pedestrians very reliably. Operating on a 24-7 schedule we can gather large trajectory data set in short times. Clearly crowds can display very complex behaviors. When moving in a crowd we constantly do our best to avoid collisions. How does this happen? In order to get a quantitative understanding we decide to first focus on an ever-simple situation one person walking alone in a corridor. What we did was to look at the motion of the pedestrian along the corridor for several months. What happens is that you can approximate the motion of anything with the pedestrian in the corridor as a trajectory on average with some fluctuations or runs which make the trajectory fluctuating along the main. The interesting point is what happens in the case where the pedestrian changes his mind along the trajectory and goes back to its original direction. The level of noise that we need for all pedestrians is compatible with these rare events and you can observe with a red line the probability that we can predict for these rare events and this is a non-trivial result. With the description of non-infecting pedestrians at hand we now can focus on more elaborate situations which can occur in crowds. In particular we are interested in how pairs avoid each other. The novelty of the train station data set allowed us to find situations wherein two pedestrians are on collision course while not interacting with people around them. Based on the custom-made interaction we are able to efficiently identify the pedestrians who meet those requirements. In the clickbait now you can see situations we capture with the solid lines we indicate interacting with the pedestrian while the dotted lines indicate when there wasn't an interaction. Pedestrians were too far apart to influence each other. To get an understanding of the avoidance interaction we defined an experiment between us. We captured the lateral distance between the two pedestrians. When we look now at how the initial distance changes towards the midway point we see that on average when the distance is small people change their path trying to avoid each other. If the initial lateral distance is significantly large people don't tend to change their course. If we now look at the second part of the interaction we see that on average people don't tend to change their trajectories anymore. To model the avoidance behavior we use a short-range radial force similar to an approach used in fluid dynamics that prevents particles occupying the same location. But contrary to molecules and particles people can navigate based on their vision. Therefore a second interacting force was introduced. With the combination of these two forces we were able to reproduce the statistics of not only a path maneuver where pedestrians modified their path when perceived the lateral distance is less than 1.5 meters. But also their speeds simulations matches the modes of walking speeds before and after the encounter. And even numbers of collisions the model agrees with measurements quantitatively how many pedestrian pairs are within 60 centimeters of each other. And so why pedestrians do not constantly collide with other pedestrians? We started to answer this question looking at simple binary collisions in dilute bi-directional flows. However, the best is yet to come. There is still so much to study and understand about the dynamics of human crowds. We are currently looking at how pedestrians behave in dense crowds, in cross-flow conditions, in generic flow settings when boarding or exiting from a train. We are also studying how pedestrians react when illumination is changed or other nudging stimuli are applied. Thanks very much for watching and don't hesitate to contact us.