Hybrid techniques for pedestrian simulations

Abstract

There is considerable interest in the simulation of systems where humans move around, for example for traffic or pedestrian simulations. Multiple models for pedestrian simulations exist: cell based models are easy to understand, fast, but consume a lot of memory once the scenario becomes larger; models based on continuous space, which are more economical with memory usage, however, use significantly more CPU cycles. 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 implies 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. This paper introduces this hybrid system, and presents performance results measured in a real-world example. © Springer-Verlag 2004.