Hybrid Control Architecture for Unsignalized Intersection Management System (CARLA/ROS)

Описание: Hybrid Control Architecture for Unsignalized Intersection Management for Multiple Autonomous Vehicles using Nonlinear MPC and Multi-Agent TD3
Eng. Ahmed Hatem
A hybrid control architecture for unsignalized intersection management for multiple autonomous vehicles is proposed. A high-level decision module based on Multi Agent TD3 (MATD3) algorithm was proposed along with a low-level control module based on Nonlinear MPC (NMPC) controller. The high-level decision-making algorithm was trained to find an optimal policy that could avoid collisions between vehicles in an intersection by choosing a specific reference velocity for each vehicle entering the intersection. The low-level controller is used to make sure that each vehicle follows its given trajectory while handling the comfort of passengers in the car by generating a smooth acceleration profile. Decoupled nonlinear longitudinal and lateral dynamic vehicle models were used to predict the future states of the systems in the NMPC algorithm. The proposed approach was tested in an intersection inside CARLA simulator, while ROS handled the communication between the different modules. The MATD3 algorithm was trained in a custom-built multi-agent intersection environment in CARLA simulator. The overall performance was found to be satisfactory.

Please, visit the research group website for more information about the academic and research activities http://mrs-lab.com/