A VERSATILE TRAFFIC SIMULATOR ABLE TO COMBINE CARTOGRAPHIC OPEN DATA WITH VARIOUS LIBRARIES OF AGENTS AND INTERACTIONS
Purpose of the project
The goal of this project is to facilitate the setup of adaptable Multi-Agent Traffic Simulations. The TrafficGen platform developed in this project is not aimed at competing with the many powerful tools developed for dedicated purposes, but instead to provide a multipurpose simulation tool that allows the designer to test several hypotheses without major code changes.
Therefore, TrafficGen focuses on the integration of existing knowledge, i.e:
- Behavioral models: various models of vehicle behavior have been designed in the literature, focusing on specific points of interest (driver psychology, conflicts management in crossroads, cabin ergonomics, changes in the road infrastructure, etc.). Through an appropriate representation of behaviors, especially by separating interactions from agents, and declarative aspects of the model from the procedural ones, existing models can be integrated in a modular way to the simulator. In TrafficGen, we use the IODA Interaction-Oriented Approach so as to ensure such a modularity.
- Open cartographic data: a lot of cartographic information have been made publicly available through participative projects such as the OpenStreetMap Project (OSM). Besides, the OpenDrive format is also widely spread in road simulation. For now both formats are supported by TrafficGen.
- Open flux information: in order to generate realistic flux of vehicles, several methods can be applied, from the use of statistical laws, to recorded data and real-time traffic information. We are currently working on the integration of such real-time data in TrafficGen.
Finally, to ensure a maximal reusability of the agents and interactions developed within TrafficGen, each entity is modeled through an agent as specified by the IODA methodology. This allows to easily change specifications regarding the level of "intelligence" or "autonomy" of entities, as well as to benefit from a homogeneous representation of all entities. Their ability to act, change or undergo actions is fully described by the interactions they can participate in, while their perception, action or cognition specificities are taken into account in the atomic primitives used by those interactions.
As a consequence of this, agents in TrafficGen are:
- "vehicles" indeed or cars, trams, cycles, pedestrians...)
- but also road infrastructure elements which interact with vehicles: crossroads, traffic lights, traffic signs...
- and also tools to setup experiments and record the simulation results: vehicles generators, probes, events dispatches (such as speed reducers operating on a given area), etc.
Each category can be extended at will depending on the needs of the experiments to run.
- A. Bonhomme, P. Mathieu, S. Picault (2015). "TrafficGen: a flexible tool for informing agent-based traffic simulations with open data", in: Y. Demazeau, K.S. Decker, J. Bajo and F. de la Prieta (eds.), Proceedings of the 13th International Conference on Practical Applications of Agents and Multi-Agent Systems (PAAMS'2015), p. 259-262, Springer LNCS 9086 (demonstration, 2nd prize IBM Award).
- A. Bonhomme, P. Mathieu, S. Picault (2014). "A versatile description framework for modeling behaviors in traffic simulations", in: G.A. Papadopoulos (ed.), Proceedings of the IEEE International Conference on Tools with Artificial Intelligence (ICTAI'2014), p. 937-944.