| 研究課題/領域番号 |
22K11918
|
|---|
| 研究種目 |
基盤研究(C)
|
| 配分区分 | 基金 |
|---|
| 応募区分 | 一般 |
|---|
| 審査区分 |
小区分60020:数理情報学関連
|
|---|
| 研究機関 | 筑波大学 |
|---|
研究代表者 |
Aranha Claus 筑波大学, システム情報系, 助教 (80629858)
|
|---|
| 研究分担者 |
BOGDANOVA ANNA 筑波大学, システム情報系, 助教 (70924463)
長谷部 浩二 筑波大学, システム情報系, 准教授 (80470045)
|
|---|
| 研究期間 (年度) |
2022-04-01 – 2025-03-31
|
| 研究課題ステータス |
交付 (2023年度)
|
| 配分額 *注記 |
4,290千円 (直接経費: 3,300千円、間接経費: 990千円)
2024年度: 1,430千円 (直接経費: 1,100千円、間接経費: 330千円)
2023年度: 1,430千円 (直接経費: 1,100千円、間接経費: 330千円)
2022年度: 1,430千円 (直接経費: 1,100千円、間接経費: 330千円)
|
|---|
| キーワード | simulator / agent based models / decision making model / user interface / unity / multi agent systems / machine learning / urban planning / natural disasters / evacuation / city generation / Simulation / Multi-Agent / Mobility / Adversarial Learning |
|---|
| 研究開始時の研究の概要 |
The planning of policies for the prevention of transmissible diseases can be improved by the use of reliable simulators. In this research we study how to use AI to create realistic multi-agent simulators that can reproduce how real people move in cities and events such as festivals and evacuations.
|
| 研究実績の概要 |
In 2023 we have developed a disease simulator which considers the evacuation of individuals during an earthquake and the intent of mask usage. This simulator was named "Koudou". The Koudou simulator was used to model the spread of Covid under different mask usage policies and earthquake evacuation, and the results of these simulations were presented at the 2023 ALIFE international conference. The Koudou simulator was published publicly on github.
We have continued the development of the Koudou simulator by designing a User interface that allows the user to investigate the state of the simulation and modify initial variables. The goal of the user interface is to allow researchers to find issues with models, and community members to learn the effects of different policies. The Interface module is being developed using the UNITY game engine to increase the public awareness of the project.
Finally, we have designed an agent model that considers the emotional state of the agent. In 2023, many people around the world stopped considering COVID an important disease, while other people still consider it a dangerous disease. To reflect this reality, the emotional state model reflects how important each agent considers COVID by taking into account news, the health level of the agent, and the overall infection levels of the community. The emotional state determines the agent's decisions regarding masking and self isolation, and can be used to simulate the evolution of a community's awareness of the disease over time.
|
| 現在までの達成度 (区分) |
現在までの達成度 (区分)
3: やや遅れている
理由
For 2023, we expected to implement a complete version of the user interface, as well as the agent emotional model, and publish these two advances on a paper at the ALIFE 2024 conference. Unfortunately, although the design of the emotion model was completed, the programming tasks related to the implementation model and the interface were not completed in time for writing the paper. We identify the reasons and fixes as follows:
- Regarding the development of the User Interface in Unity, it turns out that the programming task was more complex than expected. It was difficult to include the simulator directly in the unity engine, so we had to generate a separate program that reads the output of the original simulator and displays it. Also, the programmer invited for this job had difficulty understanding the simulator code. To avoid this delay in the future, we will be careful to create better onboarding documentation for the simulator code.
- Regarding the implementation of the emotion model, on one hand it was delayed by the delay of the User Interface development task. Also, the student who proposed and designed the emotion model did not continue into the PhD, so the transfer of knowledge also caused some delays.
|
| 今後の研究の推進方策 |
The project objectives for this year are as follows: (1) Complete the Unity Interface; (2) Implement the emotion model in the main simulator; (3) Design and Implement a mobility learning model.
- Regarding (1): We will finish the implementation of the User Interface in Unity as planned for 2023.
- Regarding (2): We will implement the designed model, and perform experiments to show how the average behavior of the agents evolves under different conditions of news and disease intensity. We will compare these results to observed situations in different cities and periods in Japan to validate the model.
- Regarding (3): Our goal is to develop the ability of the agents to learn movement strategies that are similar to patterns observed in real systems. Currently, we are investigating the use of graph neural networks to learn information from transportation graph data that exist in quantity.
|
