Congestion is an important characteristic of transportation systems and a pervasive issue that affects millions of people in their daily lives. Typically, congestion is observed on highways and roads when cars are forced to stop due to high levels of traffic, preventing them from continuing their journey.

However, congestion is not limited to roadways; it is also prevalent in pedestrian crowds, with people experiencing it regularly when riding trains during rush hours in busy urban centers. This problem is particularly pronounced in rapidly developing Asian cities such as China, Indonesia, and India, but European and American cities are also grappling with congestion-related challenges.

Measuring congestion in pedestrian crowds is a complex task because "congestion" is not a property that can be directly measured like time, distance, or speed. In the context of my research, I have developed a method that assesses the "smoothness" of pedestrian flows, enabling us to determine congestion levels (or the so-called congestion number) within crowds of people.

This method has undergone testing in various environments, allowing us to identify lane formations in bidirectional streams, differentiate between homogeneous crowds (where all people walk at the same speed) and heterogeneous crowds (where people move at different speeds), and pinpoint the riskiest areas during evacuations. It was also tested in Shinjuku station, one of the busiest railway stations in the world.

Furthermore, this method also enables us to estimate the risk associated with how people move. This information can be invaluable during mass events or for monitoring crowds in transportation hubs, aiding security personnel in taking appropriate measures when dangerous levels of risk are reached.

Bidirectional flow is a common occurrence in pedestrian traffic, such as the way crowds of people move in corridors, crosswalks, or on sidewalks/walkways. Researchers in the field of pedestrian traffic have particularly focused on bidirectional flows due to the self-organization mechanism that emerges when people arrange themselves into lanes.

One of the most interesting and fascinating aspects of lane formation is that people can organize themselves without the need for a leader or prior knowledge of how they should behave in a group. Lanes simply emerge as a collective process in which each person tries to avoid collisions with the counter-flow (you can watch a video here that illustrates the lane formation process).

However, lanes do not always appear, and sometimes people coming from both directions get stuck, forming a dangerously dense crowd. This has been a contributing factor to tragedies in the past, such as the Love Parade in Duisburg in 2010.

Researchers, in particular, have been trying to determine whether bidirectional flows are more dangerous when the number of people from both directions is equal or when there is a significant difference between them.

In my research, I thoroughly investigated bidirectional flows and the lane formation process, connecting it with past literature on the subject. Our conclusion is that lane formation is more challenging when the number of people from both directions is equal because individuals need to consider both the crowd coming from the opposite direction and the people on their left and right sides. However, lanes are more stable when the groups from both directions are similar in number, leading to increased long-term efficiency. When there is a substantial difference in the number of people between both groups, lane formation is more challenging, but interactions are relatively limited, resulting in minimal or no long-term changes.

When groups of people cross in a perpendicular direction, interactions become difficult. Think about a crowded train station with people moving straight along a corridor. If you need to cross that corridor from one side to the other, you need to check the people in front of you, but also keep an eye on the right (or left), as those walking in the corridor come from the side. This kind of interaction is more difficult compared to the “simple” bidirectional flow, in which the people you need to interact with generally come from the same direction you are walking in. In a crossflow, your goal is to move straight, but the people you need to interact with come from the side, and we are not very good at looking sideways (unlike crabs, for example, that have 360-degree vision). The typical strategy to reach the goal while minimizing collisions is to partially move with the incoming flow, deviating a bit from the goal.

This strategy, when applied by a certain number of people, creates stripes that move at an angle, which is half of the crossing streams. In other words, if two corridors cross at a 90-degree angle, stripes will form at a 45-degree angle. So far, so good. The formation of stripes is a fascinating, yet not entirely unexpected, phenomenon. However, unlike lanes in the bidirectional flow, quantifying stripes and determining the conditions under which they form is not an easy task.

In my research, along with colleagues, I performed experiments to check what the minimum crowd density required to observe stripes is, whether they occur, and to measure their angle. Experimental results were also backed by a numerical model showing properties similar to the experiments.

Assessing crowd conditions and replicating crowd movements using numerical models is undoubtedly an important task. However, ultimately, the goal is to influence and modify crowd behavior when deemed necessary.

In my research, I delved into how information provision can impact crowd behavior and which types of pedestrians are more likely to influence overall crowd performance. Both theoretical scenario studies employing numerical simulations and experiments involving real participants were used to examine this critical aspect of crowd management.

Additionally, I explored methods to enhance egress from crowded facilities by leveraging real-time information gathered through sensing devices, integrating them with simulators to optimize the overall egress process.

The results demonstrated that achieving optimal crowd control hinges on finding a balance between the quantity of information provided and the delivery methods used. Preparing steering strategies should take into account multiple scenarios. Furthermore, I found that targeting a specific category of traffic users can yield results closely resembling the impact of informing the entire crowd. In my research, providing information to people using wheelchairs was found to be beneficial for both them and the entire crowd. Similarly, other vulnerable users may benefit from targeted information, creating advantages for everyone in their vicinity.

In today's world, it has become a common practice to employ crowd simulation models when designing pedestrian infrastructures and planning mass events that accommodate a large number of people. A considerable number of commercial software options are available for this purpose, and new ones continue to be introduced and developed every year.

However, many of the models used for pedestrian simulation are primarily designed to replicate everyday conditions and do not facilitate the consideration of very dense crowds. Consequently, investigating crowd accidents or replicating extreme conditions is not feasible using conventional models.

Typically, pedestrian densities in most public spaces remain below 1 person per square meter. Movement becomes challenging when densities exceed 2 persons per square meter, but in scenarios like packed trains during rush hours, densities in the range of 6 persons per square meter are commonly observed. Several researchers have reported that in cases of crowd accidents, densities exceeded 10 persons per square meter, with some even reaching up to 15 persons per square meter.

Standard simulation models can only replicate crowds with a maximum density of approximately 5-6 persons per square meter. To enable the simulation of dense crowds, I have developed a simulation model capable of handling densities well above 10 persons per square meter, with the maximum density being 17 persons per square meter. This expansion allows for the consideration of extreme scenarios. The model's validation has been conducted using empirical data, demonstrating good agreement across various conditions.

All crowds are composed of people, but the relationships between their members are important factors determining overall behavior. Individuals behave differently from couples, and couples behave differently from families. Understanding how groups of people behave in various situations and discerning the differences from individual behavior is essential for enhancing the models used in simulations.

In the context of my research, I collaborated with various institutions, including the University of Milano-Bicocca, Okayama University, ATR Kyoto, and Kyoto University, to specifically study the behavior of dyads, which are groups composed of two persons.

While my role in this area has mostly been peripheral, mainly involving assisting in the execution of experiments and providing advice on analytical methods, these diverse collaborations have enabled me to acquire knowledge and expertise in group behavior and the methods employed to analyze and simulate their behavior.

Typically, pedestrians are detected using cameras or specialized sensors (more recently, distance sensors are increasingly used), which, although relatively expensive, also raise concerns about violating people's privacy. Although solutions that do not rely on privacy-sensitive data are under development, pedestrian spaces are more user-friendly when they don't require a large number of cameras and sensors.

In today's world, the majority of people walking in public spaces carry a smartphone or a tablet. These types of devices come equipped with a plethora of sensors and maintain constant connectivity to a communication network. While GPS is sometimes employed to determine people's positions and estimate their speeds, this method also involves handling privacy-sensitive information. A more efficient and user-friendly solution involves leveraging the inertial sensors present in electronic devices to estimate people's movement speed by analyzing their body motion.

In the context of this research, I evaluated the feasibility of using inertial sensors from commercial electronic devices to estimate the speed and density of pedestrian crowds. Our research demonstrated that under controlled conditions, both speed and density can be estimated with sufficient accuracy. However, our findings also indicated that a large-scale application may necessitate several improvements and would require a relatively substantial number of active users to ensure reliable results.

The software platform developed to collect inertial data from multiple devices in real-time was also utilized by colleagues to measure body orientation using the gyroscope sensor.