There exist multiple models for pedestrian simulations. Cell based models are easy to understand, fast, but consume a lot of memory once the scenario becomes larger. In models based on continuous space, which need almost no memory at all, however, the CPU becomes the bottleneck soon.In our project "Planning with Virtual Alpine Landscapes and Autonomous Agents", we simulate an area of 150 square kilometers, with more than thousand agents for one week. Every agent is able to move freely, adapt to the environment and make decisions during run time. This decisions are based on perception and communication with other agents.This requires a simulation model that is fast and still fits into main memory of a typical workstation. We combined the advantages of both approaches into a hybrid model. This model exploits some of the special properties of the area.• Hikers tend to walk on trails. It is possible to fit a coordinate system on a graph of these trails. Using this coordinate system, a continuous simulation is possible.• Obstacles like houses, trees, or rivers influence the route choice of hikers. We developed an algorithm which adds additional nodes to the exiting graph for each obstacle. The hiker is not only able to walk around the obstacles. but also to take the path length into account during trip planning.• Paths in the Alps are not like streets, their walkability differs a lot. The speed of the hikers is influenced by the quality of the trail. We added a grid known from cell based simulations, which allows us to control the speed of the hikers. Parts of this grid are dynamically loaded as needed.This paper will present an overview over this hybrid system, and some performance results.

BioOne Complete (complete.BioOne.org) is a full-text database of 200 subscribed and open-access titles in the biological, ecological, and environmental sciences published by nonprofit societies, associations, museums, institutions, and presses.

There is considerable interest in the simulation of systems where humans move around, for example for traffic or pedestrian simulations. Any such simulation consists of two layers: the simulation of the "physical" system, which includes effects such as interaction with other agents or the environment; and the simulation of the "mental" layer, which generates strategies of the agents. The traditional way to couple the modules is to use files. The disadvantage of that approach is twofold: The computational performance is limited by I/O; and the modules can only be run sequentially. In order to overcome these problems without sacrificing modularity, a messagebased approach is presented. Agent strategies are sent via messages to the simulation of the physical system, which executes them and sends back performance information in the form of "events". The strategic modules listen to these events, memorize them in some appropriate way, and possibly generate revised strategies. These strategies are sent to the simulation of the physical system immediately, so that the representation of the agent in the physical system will switch to the new strategy right away. In addition, the same messages can also be used to plug helper modules, such as viewers or recorders, into the system. An implementation of the framework is tested within our project, which explores the feasibility of using autonomous agent modeling to evaluate future scenarios in a tourist landscape in the Swiss Alps.

In order to accurately simulate pedestrian behaviour in complex situations, one is required to model both the physical environment and the strategic decision-making of individuals. We present a method for integrating both of these model requirements, by distributing the computational complexity across discrete modules. These modules communicate with each other via XML messages. The approach also provides considerable flexibility for changing and evolving the model. The model is explained using an example of simulating hikers in the Swiss Alps.

No abstract