Participatory Simulation

As human society is getting highly-developed and complicated, it becomes hard to design suitable social systems for the society. Social simulation has been regarded as a promising method to understand complicated human society and predict probable social problems. At the moment, it is required to reproduce the society from the microscopic view point. Therefore, micro-simulation is required rather than traditional macro-simulation. Our research objectives are to newly develop technologies and methodologies for very fine-grain scale social simulation and human modeling techniques.

Multiagent simulation has been focused as a form of micro-simulation because it is suitable to reproduce human society. For obtaining valuable knowledge/information through multiagent simulation, one of the important issues is the design of agent models and interaction models. We tackle this research issue according to our participatory technologies. We have already tried to obtain information on human behaviors using participatory simulation and construct agent model which can reproduce human behaviors. This is worldwide original approach to agent modeling for detailed/realistic social simulation.

Currently, we are working on the development of participatory technologies, modeling methodologies/techniques, multiagent simulation technologies, and software platform for mega-scale simulation.

Participatory modeling is a technology to sophisticate initially obtained agent model, which is constructed based on only literatures and expert's knowledge, through iterative RPG(Role Playing Game) or multiagent simulation with stake holders (human subjects) in each application domain.
Our modeling technology was evaluated in an actual case study of agricultural economics in the north east Thailand [Torii 06]

Participatory modeling with classification learning
In this research, when applying machine learning to the RPG log data we aim to verify the expert's hypothesis of the stakeholders' decision making model[Torii06]. For the verification process, classification learning, which extracts classification knowledge from data without prior knowledge, is more suitable than deductive learning, which regards the hypothesis as prior knowledge. The modeling process consists the following three steps:

1. Initial setting
   Create Learning data by transforming the RPG log data into a format that machine learning system can understand.
2. Elicitation of general models using machine learning system
   The models are elicited from the initial models created by experts and the data set transformed from the RPG log data. What is important here is that the expert knowledge extracted from the initial models is used in the process of the machine learning system and the learning models are as general as possible.
3. Refinement of the learned models through interactions between the machine learning system and experts
   The models outputted through the above procedure reflects expert knowledge. But it is necessary to refine the learned models using appropriate interactions between the machine learning system and experts for the models to satisfy experts and to reflect reality.

This study was organized by two research institutes (IRRI & CIRAD) in Thailand, tackling a problem of agricultural economics in the northeast Thailand.

Modeling evacuees in participatory evacuation simulation
Controlled evacuation experiments were conducted by Prof. Sugiman in 1988. He established a simple environment with human subjects to determine the effectiveness of two evacuation methods: the "Follow-direction method" and the "Follow-me method". In the former, the leader shouts out evacuation instructions and eventually moves toward the exit. In the latter, the leader tells a few of the nearest evacuees to follow him and actually proceeds to the exit without verbalizing the direction of the exit.
Considering Prof. Sugiman's real-world experiment, participatory simulations were conducted by using 3D virtual space platform "FreeWalk" and the scenario description language Q. We established a four-step process for creating scenarios: 1) defining a vocabulary, 2) describing scenarios, 3) extracting interaction patterns, and 4) integrating real and virtual experiments. This process triggers dialogue and facilitates cooperation between computer professionals and application designers.
We obtained a result close to the one obtained in a real fire drill experiment, by using the scenario description languages and the description process that we developed[Murakami 03].
We proposed a user modeling method based on participatory simulations. These simulations enable us to acquire information observed by each user in the simulation and the operating history. Using these data and the domain knowledge including known operation rules, we can generate an explanation for each behaviour. Moreover, the application of hypothetical reasoning to the generation of explanations allows us to use otherwise incompatible operation rules as domain knowledge.

Modeling Individual Drivers for Traffic Simulation
We proposed a methodology that can model each driver's individual behaviour[Tanaka 07]. The proposed approach uses a 3D virtual driving simulator to collect realistic driving log data from human subjects. We extracted driving rules from the log data by interviewing each subject. We then constructed unique driving models, which can explain each driver’s behaviour, by the application of hypothetical reasoning[Murakami 05]. We defined the driving model as a set of prioritized driving rules, each of which consists of the environment observed by the driver and the next vehicle movement in response to the observation. Each driver is expected to have a different set of rules and their priorities.

The augmented experiment combines a multiagent simulation with a small-scale experiment performed with human subjects [Ishida 07]. In the experiment, the movements of agents that simulate users are shown to human subjects in order to give them the impression that the environment is populated with a large number of users.

Evacuation Experiment in Kyoto Subway Station
We installed a disaster evacuation system that tracks passengers to help their navigation based on their current positions. Beyond conventional navigation systems, which announce route information using public loudspeakers, our system sends instructions to individuals using mobile phones. Augmented experiments are required for testing the evacuation system, because there is no other way to conduct experiments with enough reality. We placed twenty eight cameras in Kyoto station, and captured the movements of passengers in real time. As the virtual space, we used FreeWalk, a three dimensional virtual city system and used it to reproduce the passengers' behavior. Figure 1 includes a snapshot of the monitoring system; human subjects on the station platform are projected onto avatars in virtual space. A bird's-eye view of the real space is reproduced on the screen of the control center so that evacuation leaders in the center can easily monitor the experiment. As the communication channels, the leader can point at particular passengers on the screen, and talk to them through their mobile phones. When the monitor detects pointing operations, a wireless connection is immediately activated between the control center and the indicated passengers.

Experiment of Personal Evacuation Navigation with GPS Mobile Phones in Kyoto City
To develop a navigation service for humans with mobile terminals, it is necessary to test the system with a large number of human subjects. However, any experiment in the real world with many human is too expensive and rather dangerous. Thus we took the augmented experiment approach which uses a multiagent simulation to expand a real-world experiment with a few subjects.
We produced a city-wide evacuation guide system assuming the use of GPS-capable cellular phones. In this system, each user receives updated maps and instructions from his/her own guide agent. We subjected this system to an augmented experiment.
In the augmented experiment, the positions and movements of simulated users was shown on the navigation maps sent to each subject. Interviews of the human subjects confirmed that the system successfully gave the impression to the human subjects that they were participating with a large number of users. The results of this experiment give some indication of the possibility of using augmented experiments to refine city-wide navigation services.

Scenario Description Language Q
Q is a language for describing interaction between agents and humans based on agent external roles[Ishida 07]. Q does not depend on agent internal mechanisms; its goal is to describe how scenario writers should be able to request agents to behave.
The change of focus from agent internal mechanisms to interaction scenarios significantly impacts the language syntax and semantics. For example, if an agent accepts only two requests, "on" and "off," Q allows scenario writers to use just two commands, "on" and "off." This does not mean that the agent is not intelligent, only that the agent is not controllable once it is turned on.
Q has been combined with several legacy agent systems: Microsoft Agent, FreeWalk, CORMAS, and Caribbean. For example, you can control several hundreds of agents in 3D virtual city simulator FreeWalk/Q on a PC. FreeWalk/Q was used for evacuation experiment at Kyoto subway station, simulation of fire disaster at Hotel New Japan in 1982 (under the collaboration with National Research Institue of Fire and Disaster) and psychological experiment in virtual society (collaboration with Stanford University). Q also contribute to establish a bridge between agent designers (computer professionals) and application designers (scenario writers). To maximize this benefit, we introduced IPC (Interaction Pattern Card) with Excel interface. We expect that effective dialog emerges from the interplay between the two different perspectives during the process of formalizing interaction patterns. We also introduced a tool to generate Q codes from Visio finite state machine diaglams. For more detail, Please visit Q web site.

Large-scale Multiagent Platform Caribbean/Q
The spread of mobile terminals will realize a ubiquitous environment for city-dwellers. Using this environment, we can build large-scale societal information systems. Multiagent simulations are promising approach to evaluate these systems.
In developing a large-scale societal information system, experts of the intended domain will design the agent interaction protocols while computer experts will develop the agent system. In large-scale societal information systems, each agent faces a variety of situations. Furthermore, systems have to manage a large number of agents that model human behaviors. For these issues, we developed and evaluated a large-scale multiagent platform where the execution of agents scenario and the implementation of agents are explicitly separated[Nakajima 06].
We also developed a participatory simulation system for evacuation guidance experiments on this platform. Using this system, a commander can participate in a simulation through transcendence interface. Other subjects participate in the simulation as evacuees by walking real-world. Our system gets positions of subjects by GPS-capable cellular phones and reflects them to the simulation.
For more detail about Caribbean/Q, please visit mega-navigation project web site.

3D Virtual City Space Simulator FreeWalk/Q
FreeWalk is a platform where human participants and autonomous characters can socially interact with one another in a virtual city space[Nakanishi 04].
FreeWalk has such facilities for large-scale multiagent simualtions as voice communication, speech recognition, speech synthesis, 3D human behavior visualization, dynamics computation for agents' walk and distributed processing on PC cluster.
For more detail about FreeWalk, please visit FreeWalk web site.

Toru ISHIDA (Professor)
Hiromitsu HATTORI (Assistant Professor)
Yuu NAKAJIMA (Assistant Professor)
Jie ZHOU(M2)
Hiroaki KINGETSU(M1)
Keita UEDA(B4)
Akimasa NAKASHIMA(B4)

[Contact]
Hiromitsu HATTORI

[Conferences]

[Ishida07] Toru Ishida, Yuu Nakajima, Yohei Murakami and Hideyuki Nakanishi. Augmented Experiment: Participatory Design with Multiagent Simulation. International Joint Conference on Artificial Intelligence (IJCAI-07), 2007. (pdf, 597KB)

[Torii06] Daisuke Torii, Toru Ishida and Francois Bousquet. Modeling Agents and Interactions in Agricultural Economics. International Joint Conference on Autonomous Agents and Multiagent Systems (AAMAS-06), pp. 81-88, 2006. (pdf, 544KB)

[Murakami05] Yohei Murakami, Yuki Sugimoto and Toru Ishida. Modeling Human behaviour for Virtual Training Systems. The 20th National Conference on Artificial Intelligence (AAAI-05), pp. 127-132, 2005. (pdf, 374KB)

[Nakanishi04] Hideyuki Nakanishi, Satoshi Koizumi, Toru Ishida and Hideaki Ito. Transcendent Communication: Location-Based Guidance for Large-Scale Public Spaces. International Conference on Human Factors in Computing Systems (CHI-04), pp. 655-662, 2004. (pdf, 801KB)

[Murakami03] Yohei Murakami, Toru Ishida, Tomoyuki Kawasoe and Reiko Hishiyama. Scenario Description for Multi-Agent Simulation. International Joint Conference on Autonomous Agents and Multiagent Systems (AAMAS-03), pp. 369-376, 2003. (pdf, 1.32MB)

[Nakajima06] Yuu Nakajima, Hironori Shiina, Shohei Yamane, Hirofumi Yamaki and Toru Ishida. Caribbean/Q: A Massively Multi-Agent Platform with Scenario Description, International Conference on Semantics, Knowledge and Grid (SKG-06), 2006. (pdf, 2.82MB)

[Symposiums and Workshops]

[Tanaka07] Yusuke Tanaka, Yuu Nakajima and Toru Ishida. Learning Driver’s Model Using Hypothetical Reasoning. Pacific Rim International Workshop on Multi-Agents (PRIMA-07), Lecture Notes in Artificial Intelligence, Springer-Verlag, 2008. (pdf, 349KB)

[Journals]

[Ishida02] Toru Ishida. Q: A Scenario Description Language for Interactive Agents. IEEE Computer, Vol.35, No. 11, pp. 42-47, 2002. (pdf, 617KB)