Skip to Main content Skip to Navigation
New interface

Systèmes multi-agents, auto-organisation et contrôle par apprentissage constructiviste pour la modélisation et la régulation dans les systèmes coopératifs de trafic

Maxime Guériau 1, 2 
Abstract : In a near future, connected and automated vehicles will progressively replace current vehicles, leading to deep changes in transportation. The driver will be soon assisted and then replaced by an embedded system, able to act quicker, relying on a more robust and precise representation of its surrounding environment. However, some steps are still needed before coming up with such a level of automation since the vehicle environment is complex and unpredictable. This work intends to anticipate the introduction of these new kinds of vehicles by providing cooperative behaviors at both infrastructure and vehicle levels, at the same time allowing a decentralized control of these systems. We propose a distributed modeling framework, using multi-agent systems, relying on the coupling of the system dynamics\string: information, communication and reliability (modeled through the concept of trust). The next step was to develop a simulation framework enabling the implementation of our models for connected vehicles applications. We present a new microscopic traffic simulator, built as an extension of an existing platform, and able to model information exchanges using messages between vehicles and with the infrastructure. All data are provided by sensors and all entities, modeled as agents, are autonomous regarding their decision process. Thanks to the simulator, it is possible to imagine new control strategies relying on recommendations disseminated by the connected infrastructure. Consistency and interdependence of the simulator components are ensured by the dynamic coupling. As for the vehicles' dynamics, we propose a bilateral multi-anticipative model that integrates additional information from communications in the vehicle decision process. Results in simulation confirm that the model is able to reduce the propagation of perturbation through the flow, leading to a more homogeneous and stable traffic. One of the major issues regarding traffic control strategies will be to dynamically adapt the action policy to the several deployment stages of cooperative transportation systems. The similarities with Artificial Intelligence problems like cognition motivate a more abstract study\string: how to model an autonomous system able to control its environment. We choose the constructivist approaches, that propose to model the cognition process as an iterative building process. For cooperative traffic, the benefits lie in the ability of the system to generate its own strategies, relying or not on domain specific knowledge, and then make them evolve to be adapted to vehicles in the flow. The results from our approach are presented in two distinct simulation frameworks. The first one is an experimentation prototype aiming at highlighting the low-level behaviors in a simplified environment. In this context, we show that the model is able to combine efficiently several individual concurrent representations in order to build a high-level representation that can be adapted to several contexts. The second framework is the traffic simulator where the results lead to some insights about the potential of our approach for such realistic applications.
Complete list of metadata

Cited literature [218 references]  Display  Hide  Download
Contributor : Maxime Guériau Connect in order to contact the contributor
Submitted on : Monday, April 10, 2017 - 10:12:54 AM
Last modification on : Friday, September 30, 2022 - 11:34:16 AM
Long-term archiving on: : Tuesday, July 11, 2017 - 12:27:36 PM


Distributed under a Creative Commons Attribution - NonCommercial - NoDerivatives 4.0 International License


  • HAL Id : tel-01504421, version 1


Maxime Guériau. Systèmes multi-agents, auto-organisation et contrôle par apprentissage constructiviste pour la modélisation et la régulation dans les systèmes coopératifs de trafic. Intelligence artificielle [cs.AI]. Université de Lyon I Claude Bernard, 2016. Français. ⟨NNT : 2016LYSE1318⟩. ⟨tel-01504421⟩



Record views


Files downloads