Modeling of traffic simulation on driving simulators
Context and motivations
The Technical Center for Simulation of Renault has been developing and using driving simulators for more than ten years. These simulators are used for example for experiments in ergonomics of the driver’s cab, validation of embedded systems, comfort, or for the design and the validation of car lightings. The use of these tools is fully integrated in the vehicles development processes, and so an as realistic as possible reconstitution of the driving environment is crucial.
All Renault’s and Nissan’s driving simulators use the SCANeR© software, developed by the Technical Center of Simulation. This includes the module managing the traffic, module which has to allow the setting up of experimental protocols corresponding to accurate and controlled situations. For example, the implementation of a hard braking situation should be easy and reproducible. Although the current traffic module is successfully used for number of experiments, our users have today reached some of the limits of its design’s paradigms.
For example, the different vehicles behave all in a similar way, even if they are trucks or bicycles, driven by a responsible person or a drunken one. Likewise, the behavior of the driver is governed by an automaton, which does not allow to simulate all subtleties of the decision-making process of a real driver. The individual behavior of each vehicle is regulated through a number of limited, interdependent parameters, empirically adjusted on a set of test cases. The global behavior of the traffic widely depends on these adjustments, without offering an explicit way to parameterize the settings.
 Visual output on urban environment |
|
Objectives
The aim of this thesis will consist in developing a traffic module interfaced with SCANeR© III, which solves the most limiting problems for the experiments.
First, we have to be able to simulate various typologies of drivers, using different driving strategies, while offering simple but accurate behavior’s adjustment parameters. Ideally an exhaustive set of intuitive, independent and self-coherent parameters would be available – whatever the selected set of parameters the resulting behavior remains credible –, and an automatic learning method for possible nonintuitive or driver’s nonintrinsic parameters could be provided. This parameter setting will be based on statistical data measured in real situations, which gathering could be part of the research undertaken during the thesis. In addition, the model will have to be scalable enough to limit the difficulties of integration of new behavioral models.
It is also crucial to be able to control the traffic through scenario elements. In order to set up particular driving situations, it is indeed necessary to be able to force the virtual vehicles to overtake, to hard brake… It will be important to make sure that these controls do not detract too much the global realism of the simulation. Finally, it is important to be able to make it possible to simulate various types of vehicle by taking into account their physical characteristics.
Skills in computer science, algorithmic (in artificial intelligence for example) and behavioral modeling will therefore be needed for the study.
Course of the thesis
The thesis is co-directed by Dr. Andras Kemeny, chief of the Technical Center of Simulation of Renault , and the Pr. Philippe Mathieu, in charge of the Behaviours and Multi-Agent Systems team of the Computer Science Laboratory of Lille . The main directions of research will be:
– Bibliographical study: practical and bibliographical study on the traffic and the existing traffic generators (driving simulation, city planning, traffic prevision…) and on the usable technics (artificial intelligence, multi-agents systems, fuzzy logic…), in order to have an in-depth knowledge of the various fields, which will be studied during the thesis.
– Theoretical study: proposition of a theoretical approach answering to the microscopic and macroscopic problematics of the traffic, and offering adapted adjustment capacities.
– Development: development of a real time traffic generator interfaced with the SCANeR© software.
References
[1] Alexis Champion, Alexandre Heidet and Andras Kemeny, Traffic Generation with the SCANeR© II simulator: towards a multi-agent Architecture, in Proceedings of DSC’99 Driving Simulation Conference, 1999, pp. 311-324
[2] Marcos Fernandez, Inmaculada Coma, Gregorio Martin and Salvador Bayarri, A Hierarchical Object Oriented Architecture for the Management of Complex Driving Simulation Scenarios, in Proceedings of DSC’99 Driving Simulation Conference, 1999, pp. 325-338
[3] Omar Ahmad, Yannis E. Papelis, Matt Schikore and Ginger Watson, An autonomous driver model and graphical authoring of scenarios: advanced traffic simulation tools for the NADS simulator, in Proceedings of DSC’00 Driving Simulation Conference, 2000, pp. 205-224
[4] A.D. Dumbuya, R.L. Wood, T.J. Gordon and P.D. Thomas, An agent-based traffic simulation framework to model intelligent virtual driver behaviour, in Proceedings of DSC’02 Driving Simulation Conference, 2002, pp. 363-374
[5] Peter Hidas, Modelling Vehicle Interactions in Microscopic Simulation of Merging and Weaving, in Transportation Research Part C
[6] S. Donikian and B. Arnaldi, Complexity and concurrency for behavioral animation and simulation, in Proceeding of the 5th Eurographics Workshops on animation and simulation Oslo September, 1994, pp 101-113
Driving Simulation & Virtual Reality 