AVeSi: Agent-based Traffic Simulation

Simulation of realistic traffic conditions using individual agents capable of showing common misbehavior in traffic (e.g. running a red light, double parking, etc.).

Traffic simulations are currently used for road planning or optimizing evacuation scenarios, but also in video games and other virtual environments. Current systems are either completely rule based or change certain aspects at random. Dangerous situationsĀ are mostly avoided. However, such situations are necessary to recreate real-world traffic as traffic participants will occasionally violate traffic rules intentionally or unintentionally. Examples for common misbehavior are speeding, double parking, parking violation, or tailgating. These deviations from strictly rule-based behavior also depend on the type of driver and other influences (e.g. personality, road conditions, etc.). If such misbehavior is indeed part of the simulation, it is usually realized using scripted events, making the created scenarios inflexible and predictable.

The goal of this project is to develop a realistic traffic simulation within the realm of road safety education. To achieve this challenging task, a bicycle simulator (FIVIS project) will be extended with a global traffic simulation that includes multi-modal traffic participants (agents). The agents should be given personality profiles and their behavior should be driven by a model of relevant cognitive processes used by humans in traffic (e.g. perception, anticipation, decision making, etc.). An important aspect will be the modeling of individual risk propensities, leading to varying behavior in similar situations. Interesting subtasks will include, maintaining a large number of persistent (but less realistic) agents as well as a few realistic agents around a user in real-time, implementing a suitable representation of a road network, modeling the cognitive processes and more. Research topics may include, but are not limited to, path finding, multi-agent systems, graph theory, GPGPU, psychology, and 3D visualization.