7 June 2009 in Autonomous Agents and Multi-Agent Systems
23 September 2015 in ACM Transactions on Interactive Intelligent Systems
This article presents a review and a classification of interactive optimization methods. These interactive methods are used for solving optimization problems. The interaction with an end user or decision maker aims at improving the efficiency of the optimization procedure, enriching the optimization model, or informing the user regarding the solutions proposed by the optimization system. First, we present the challenges of using optimization methods as a tool for supporting decision making, and we justify the integration of the user in the optimization process. This integration is generally achieved via a dynamic interaction between the user and the system. Next, the different classes of interactive optimization approaches are presented. This detailed review includes trial and error, interactive reoptimization, interactive multiobjective optimization, interactive evolutionary algorithms, human-guided search, and other approaches that are less well covered in the research literature. On the basis of this review, we propose a classification that aims to better describe and compare interaction mechanisms. This classification offers two complementary views on interactive optimization methods. The first perspective focuses on the user’s contribution to the optimization process, and the second concerns the components of interactive optimization systems. Finally, on the basis of this review and classification, we identify some open issues and potential perspectives for interactive optimization methods.
1 November 2008 in Simulation Modelling Practice and Theory
16 March 2023 in IEEE Transactions on Intelligent Transportation Systems
Intersections are major bottlenecks for road traffic, as well as the origin of many accidents. Efficient management of traffic at intersections is required to ensure both safety and efficiency. Yet, the traditional solutions (static signs, traffic lights) are limited in their efficiency as they consider the flow of vehicles and not the vehicles at the microscopic level. By using inter-vehicular communication of connected vehicles, recent works have shown the possibility to have a great increase in the number of evacuated vehicles thanks to the possibility to give an individual right-of-way directly to each vehicle. In this context of intersections of cooperative vehicles, the scheduling of this right-of-way in order to maximize the throughput of the intersection is still a challenging task, with regard to the hybrid and dynamic aspects of the problem. In this paper, we propose an approach based on Deep Reinforcement Learning (DRL) to efficiently distribute the right-of-way to each vehicle. A Markov Decision Process model of intersections of cooperative vehicles, enabling the application of DRL, is proposed. The performance of the DRL-based scheduling is then compared with classic traffic lights, and with two state-of-the-art cooperative scheduling policies, showing the benefits of the approach (increase of the flow, reduction of CO2 emissions).