Distributed Control and Learning of Connected and Autonomous Vehicles Approaching and Departing Signalized Intersections
Semester of Graduation
Master of Science in Electrical and Computer Engineering (MSECE)
This thesis outlines methods for achieving energy-optimal control policies for autonomous vehicles approaching and departing a signalized traffic intersection. Connected and autonomous vehicle technology has gained wide interest from both research institutions and government agencies because it offers immense promise in advancing efficient energy usage and abating hazards that beset the current transportation system. Energy minimization is itself crucial in reducing the greenhouse emissions from fossil-fuel-powered vehicles and extending the battery life of electric vehicles which are presently the major alternative to fossil-fuel-powered vehicles. Two major forms of fuel minimization are studied. First, the eco-driving problem is solved for a vehicle approaching a traffic signal intersection using the deep reinforcement learning approach. The task is to find the optimal control input to the vehicle approaching a signalized intersection given the traffic signal pattern. It is assumed that the vehicle is made aware of the traffic signal through vehicle-to-vehicle and vehicle-to-infrastructure communication. A microscopic fuel-consumption model is considered. The system model, system constraints, and fuel consumption model are translated to the reinforcement learning framework. The model is then trained and simulations are presented. Practical deployment considerations are also discussed. Next, the multi-agent vehicle platooning control is considered. Vehicle platooning exploits the aerodynamics of vehicles that follow each other closely in a line to reduce the total energy consumption of the vehicle fleet. Graph-theoretic methods that characterize the interaction of multi-agents are studied using matrix-weighted graphs. Particularly, the roles of the matrix weight elements in matrix-weighted consensus are examined and the results are demonstrated on a network of three agents. The results are applied in vehicle platoon splitting and merging for a vehicle approaching a traffic stop.
Ogbebor, Joshua Onyeka, "Distributed Control and Learning of Connected and Autonomous Vehicles Approaching and Departing Signalized Intersections" (2022). LSU Master's Theses. 5517.
Meng, Xiangyu. Wei, Shuangqing. Gu, Guoxiang