Agents on Wave Field: Simulation of Traffic on Train Station H. Mamiya M. Kubo H. Satoh A. Namatame Dep. of Computer Science Dep. of Computer Science Dep. of Computer Science Dep. of Computer Science National Defense Academy National Defense Academy National Defense Academy National Defense Academy Yokosuka,JAPAN Yokosuka,JAPAN Yokosuka,JAPAN Yokosuka,JAPAN yonse@mb.infoweb.ne.jp masaok@nda.ac.jp hsato@nda.ac.jp nama@nda.ac.jp Abstract In this paper, we would like to simulate typical 2 behaviour of people in traffic at train station. To avoid combinatorial problem, we adopt VPM architecture. An information is represented by progressive wave on traditional wave field. The Agent has virtual sensors by FFT and retrieves information from the wave field. Also, agent can propagate information as like as sing a song. Our simulation over1500 agents work on personal computer. I. Introduction In this paper, we would like to simulate traffic of people because the behaviour is very complex, very dynamic, and interesting[1]. Usually, people are influenced by a lot of topological information from far and near, it needs large calculation cost that is grown up exponentially relative to the number of object because people interact each other[2][3]. To avoid this problem, Aoki adopts that the effect of information at far is pre-defined [4]. It seems to be reasonable for their aim because the elegant simulation is verified by actual measurement. However, the algorithm to set appropriate basic flow of people is not described. We adopt the following assumption to avoid this combinatorial problem. 1) Agents (includes objects) that are far are not interact directly. 2) Each agent transmits information to field. 3) The information is propagating on the field. 4) Each agent observes field condition around near by its virtual sensors. Namely, it needs propagating time for agent to influence other agents and delay is occurred infallibly. It is a strong assumption but the profit is incredible because designer just make an agent that reacts to neighbouring states of field. Needless to say, it depends on aim of simulation that the assumption derives the above profit. But it is difficult to classify simulation aim and load of designer, in this paper, we would like to show applicable simulations, typical 2 behaviour of people in traffic at train station. The remain of this paper is composed as follows. First, we introduce VPF [7] to construct the field that agents propagate their information. Next, the virtual sensor is explained. The passenger agent algorithm is proposed at section IID. In section III, we show 2 simulations at typical situation at train station. 1 2 Agent On Wave Field Passengers in train station are attracted by other passengers, guidance indicator, commercial messages, and train. These information sources are spread in a space. Their positions are changeable in sometimes. To realise the above information space (based on affordance theory [5]), Yokoi proposed Vibrating Potential Field (VPF) architecture. In his architecture, an information is defined as a vibration source and the information is propagating on wave field (See Fig.1). The agent can receive the information via the wave field by wave analyser as like as human receive by using eye, ear, nose, and skin and react these information. The agent on VPF is able to design by set of simple Living Things observes its environment by its sensors.(e.g. Eye,Ear) Acoustic and Optic information propagating among Air Living Things Object Wave propagating on ∂ 2 φ ∂ t 2 = a ⋅ ∂ 2φ ∂ x 2 + ∂ 2φ ∂ y 2       Virtual Sensor in Simulator Agent Object Figure 1 Policy of VPF simulator rules. By superposition principle of waves, the agent on VPF architecture can know information about information source far away only by observation its neighbourhood (See Fig.2). In VPF, differences among information type, for example, sound and light corresponds with characteristics of