Desenvolvimento e avaliação de um sistema multiagente aplicado à assistência ao condutor em interseções urbanas

Abstract

Urban intersections are critical points in traffic due to the confluence of vehicle flows and high risk of accidents. Technological tools, such as traffic simulations, data science, and artificial intelligence, are essential for analyzing patterns, preventing accidents, and optimizing vehicular flow. The main objective of this study is to design and evaluate a multi-agent system applied to driver assistance, focusing on safety at urban intersections, where the risks are more pronounced. To achieve this goal, the study used the Java Agent DEvelopment Framework (JADE), which enables the modeling of autonomous agents that interact in a simulated environment. The Simulation of Urban MObility (SUMO) is used to provide a detailed simulation of traffic interactions in complex urban scenarios. The integration between JADE and SUMO is facilitated by the Traffic Simulation Manager Application Programming Interface (TraSMAPI), an interface that allows you to coordinate and synchronize communication between the two platforms. The results showed the potential of multi-agent systems in the prevention of accidents at urban intersections. The cooperative messaging protocol ACL (Agent Communication Language) eliminates the collisions recorded in the purely reactive SUMO scenario, while imposing an average increase of 7.3% in travel time and only 0.18s of waiting time per vehicle. These results suggest that multi-agent coordination can promote safer decisions without significantly compromising traffic flow. Although simulations still need validation in higher density networks and comparison with empirical data, this work contributes to the development of more resilient driver assistance systems with the adoption of an automated test protocol that varies systematically, pointing the way for applications in smart cities and for future extensions involving distributed machine learning and field calibration.