Investigating the Technical Possibilities of Real-time Interaction with Simulations of Mobile Intelligent Particles

Abstract

Using simulation systems for transport planning has become more popular in the past two decades. Most of these systems rely on the ”four-step process”. A more recent approach is to use transport-planning systems based on multiagent simulations. MATSim is a framework for building multi-agent-based simulations for transport systems. Unlike the flow-based models such as the one used by the software VISUM, multi-agent systems are based on distinct daily plans for the whole simulated population. These plans are then executed in a simulated world for all agents in parallel. One advantage of multi-agent simulations is that it is possible to build a complex system of traffic interaction by modeling the agent’s behavior with simple rules. The complex behavior of the system is broken down to the behavior of simple agents. By using the computationally inexpensive ”Queue Model”, the MATSim framework is capable of simulating even large-scale networks and population groups. The traffic demand of cities, counties or even entire countries can be simulated. For example, all of traffic in the Berlin-Brandenburg region in Germany with its seven million agents has been simulated using the MATSim framework. Unfortunately, only more or less aggregated data can be used as an output of these simulation runs. Given the enormous amount of data generated during a simulation run, it is not feasible to examine the unaggregated data. However, since an aggregated view of the results of the simulation may obscure important details, a more finely granular view would be beneficial. The aim of this thesis is to enable the researcher to have an unaggregated view of the simulation. In the first part of this thesis, ways of visualizing the unaggregated data will be examined. The unaggregated view of the actual traffic flow promises to be a useful tool when looking for causes of observed phenomena. The graphics capabilities of computers have continuously increased over the past years. Trying to visualize hundreds of thousands of agents simultaneously on screen no longer seems far-fetched. A software architecture will be developed that is capable of displaying these quantities of agents while being open and expendable enough to allow the researcher to add any form of visualization him or herself. This architecture will enable the researcher to investigate cause-and-effect chains in a multi-agent simulation by examining the unaggregated data. To achieve this, modern hardware acceleration of 3-D graphics is extended to accelerate the display of primarily 2-D traffic data. The second part of this dissertation will address possible ways to increase the execution speed of traffic simulations by means of modern graphics hardware. Accelerating the execution speed of simulations has already been tried a few times. This has normally involved incorporating large clusters of computers or using some other sort of expensive supercomputing hardware. However this is probably not a feasible way to reach a state in which ordinary engineering offices can use a multi-agent simulation to preview the results of a planned measure. Using modern graphics hardware, on the other hand, is simple and cheap. All that needs to be acquired is a e300 graphics card and a regular office PC. It will be shown that it is possible to achieve a speedup of up to 70 times more than the ”usual” Java version of mobility simulations by using graphics hardware to calculate the simulation. This will make high-performance computing affordable even for small companies. Finally, fast simulation and visualization will be brought together to create a system capable of displaying a simulation of hundreds of thousands of agents in real time.