Simulation-based Social Decision Support

Analyzing and designing real-world social systems or planning countermeasures against various social issues is a challenging task. This is because society is a complex system comprising a variety of interactions between individual elements that can significantly alter the whole context. While many social system theories have been proposed to aid in analysis, design, and problem-solving, highly abstracted theories alone may struggle to fully capture the complex details and dynamic behaviors of real-world societies. And although social experiments are important, repeating them is often impractical due to time, cost, and safety considerations. Therefore, our laboratory tries to develop novel methods of social simulation and to apply them to address social issues.

To effectively utilize simulation for social analysis, design, and problem-solving, it is essential to understand and model the complex dynamics of social phenomena and to make highly accurate predictions based on that. In our laboratory, we are (1) throughly investigating of the characteristics of the target system, (2) modeling the behavior of the people as the essence of the target system at an appropriate resolution, (3) constructing and advancing simulations that are suitable for the problems to be solved, and (4) trying to apply them to solve real-world problems.

When introducing new policies or technologies into the current social system, it is essential to carefully consider whether they will be truly effective and whether unintended risks will arise. While quantitative predictions may be possible with a simulator that accurately reproduces reality, we must understand that the quantitative prediction does not directly translate into effective social decision-making. It is the stakeholders involved in the social system who can change it, not social simulators. As simulation researchers, our goal is to support decision-making by presenting quantitative simulation results in stakeholder consensus-building forums. If a consensus-based scenario is identified through this process, it can be implemented in society, leading to the realization of the agreed-upon social state.

Social System Modeling

When attempting to understand a complex system, such as society, there are instances where the approach of decomposing the system into its constituent elements and observing those elements to gain a comprehensive understanding of the whole context is ineffective. Humans are components of society, organs are components of a human, cells are components of an organ, and so on, ultimately reaching the world of elementary particles. However, even if we could understand the behavior of elementary particles, we would still not be able to understand the overall nature of society. In this way, systems in which the global properties or behavior arise from the local interactions of their components, but the global properties or behavior are not apparent from the individual components, are called complex systems. Among complex systems, those in which the individual components adapt their behavior to their surroundings or the whole situation are called complex adaptive systems. In our laboratory, we are challenging the modeling of complex social systems, including complex adaptive systems, using network models, cellular automata, and multi-agent systems, etc.

Social simulation models can be classified into three categories: abstract models, middle-range models, and facsimile models, in order of ascending resolution. Abstract models have simple structures, resulting in low computational costs, and therefore allow for relatively wide-area simulations even in situations with limited computational resources. Facsimile models faithfully reproduce specific social phenomena under limited conditions; however, they require substantial computational resources and are more challenging to use for wide-area simulations compared to abstract models. Abstract models are suitable for understanding social phenomena because they eliminate unnecessary elements, allowing for a clearer understanding of the underlying concepts. In contrast, facsimile models are suitable for prediction and design because they can incorporate various elements of the real world. To accurately predict various social phenomena and support social decision-making, it is desirable to realize large-scale simulations using facsimile models. Based on this concept, we are conducting research on the application of high-performance computing technologies, such as parallel computing, to enhance simulation efficiency and alleviate computational resource constraints. On the other hand, we are exploring the hybridization of abstract models and facsimile models to enable the selective application of high-resolution models to specific areas of interest.

Development & Application of Microscopic Traffic Simulator

We have been developing a multi-agent traffic simulator, "ADVENTURE_Mates," that has both large-scale and high-resolution features and can be applied to the multifaceted analysis and prediction of traffic phenomena. One of the key features of ADVENTURE_Mates is to model human intelligent behavior without overly simplifying traffic phenomena.

Vehicles implemented as intelligent agents in ADVENTURE_Mates autonomously perceive their surrounding environment and determine the next action to take. Actions are interactions with the environment, and since other vehicle agents also interact with the environment simultaneously, vehicle-to-vehicle interactions occur through the environment. While these interactions are generally local phenomena, the total sum of them can evolve into complex, large-scale phenomena, such as traffic congestion.

Developing advanced simulation models is a key approach in our research. At the same time, we are applying these models to scenarios that reflect real-world traffic systems in order to mitigate and resolve current and future traffic-related issues.

Hybrid Traffic Model

When focusing on the formation and resolution of traffic congestion, which is a typical traffic phenomenon, in a free flow state with low vehicle density or a steady congested flow state with sufficiently high vehicle density, the fluctuations of individual vehicle behaviors do not significantly affect the overall traffic flow. However, in the transition area between free flow and congested flow, or between congested flow and free flow, individual vehicle behaviors can potentially accelerate or delay the formation or resolution of congestion. By adopting a high-resolution facsimile model in the transition area and an abstract model in free flow and steady congested flow states, we have developed a hybrid traffic model that maintains accuracy while improving computational efficiency. The model is utilized to support the design of traffic systems.

Pedestrian Traffic & Mixed Traffic

The main players in road traffic are not just automobiles. When considering the redesign of road space, traffic safety, and evacuation in the event of a disaster, it is necessary to take pedestrian traffic into account. Public transportation transfers are also considered pedestrian traffic. We are working on crowd simulation by utilizing existing models, such as the Social Force Models, and proposing a new model, such as the Extended One-dimensional Pedestrian Model, that balances calculation accuracy and efficiency. Additionally, by combining pedestrian models, vehicle models, and tram models derived from vehicle models, we have realized pedestrian-vehicle mixed traffic simulation and pedestrian-vehicle-tram mixed traffic simulation. In collaboration with local governments, we are contributing to the redesign of urban spaces.

Household & Urban Dynamics

In cities, a certain structure emerges through the interaction between transportation (the short-term movement of people) and migration (the long-term movement of people). We will construct simulators that represent not only traffic flows but also household and urban dynamics, and use them to evaluate urban planning and analyze energy consumption trends. Contributing to the realization of smart cities is also one of our objectives.

To express household dynamics through simulation, we proposed a multi-agent household dynamics simulation in which households are modeled as agents. The status of residents is expressed as the properties of a household, which is transitioned probabilistically at each step. By providing real data such as birth rates, marriage rates, and migration rates, it is possible to accurately reproduce the changes in the number of households in a city and the composition ratio of household types (single-person households, households with married couples and children, three-generation households, etc.).

A residential location choice simulation in which household agents within a city choose their place of residence based on the influence of land parcels, prices, and attractiveness. The attractiveness of land parcels is defined by accessibility to urban facilities such as schools and hospitals. For example, households with children prioritize accessibility to schools, and accessibility differs between households with and without private vehicles, reflecting the differing characteristics of these household agents. As a result, the simulation can express the characteristics of districts, such as those suitable for three-generation households or single-person households.