INTELLIGENT INTERSECTION MANAGEMENT THROUGH GRADIENT-BASED MULTI-AGENT COORDINATION OF TRAFFIC LIGHTS AND VEHICLES

Loading...
Thumbnail Image

Publication or External Link

Date

2021

Citation

Abstract

This dissertation examines the problem of coordinating two different types of actors in a vehicular traffic network system, namely: the traffic lights and the connected and automated vehicles traversing the traffic network. The work is motivated by an extensive previous literature showing that traffic network synchronization has substantial potential throughput and fuel economy benefits. The literature presents many algorithms for synchronizing the traversal of intersections by connected and automated vehicles (CAVs), as well as the synchronization of traffic lights within a given network. However, the integrated solution of these two synchronization problems remains relatively unexplored. The main challenge of any algorithm proposed in this area consists of managing the trade-off between computational efficiency, communication requirements, and performance.

This dissertation seeks to contribute to the list of proposed coordination strategies for CAVs and smart traffic lights by formulating a decentralized framework based on combining ideas from gradient-based multi-agent control, trajectory planning and control barrier functions. The overall proposed control framework consists of describing vehicles and traffic lights by an extra state that directly or indirectly represent its timing (i.e arrival time for the vehicles, and switching time for the traffic lights). This timing variable evolves according to a networked multi-agent system, where the planned timing of neighboring agents governs the evolution of the planned timing of the ego agent. The planned timing state is then translated into a control action for the agents (i.e. acceleration for the vehicles, switching actuation for the traffic lights), through trajectory planning and safety regulation.

The proposed coordination framework (i) can coordinate both vehicles and traffic lights, (ii) scales efficiently to large numbers of vehicles and intersections, (iii) is computationally efficient, (iv) can work under different levels of connectivity assumptions and in the presence of human drivers, and (v) can allow for different types of coordination strategies encoded in the underlying ETFs.

Notes

Rights