Wang, Tong (Xi'an University of Technology), Pang, Hui (Xi’an University of Technology), Zhao, Bin (Xi’an University of Technology), Zhang, Ke (Xi'an University of Technology), Liang, Ning (Xi’an University of Technology)

Multi-Agent DDPG and Its Application to Longitudinal and Lateral Control of Unmanned Autonomous Car Queue

Scheduled for presentation during the Regular session "Powertrain Control II" (FrB3), Friday, November 1, 2024, 11:50−12:10, Room T3

7th IFAC Conference on Engine and Powertrain Control, Simulation and Modeling, Oct 30 - Nov 1, 2024, Dalian, China

This information is tentative and subject to change. Compiled on January 2, 2025

Abstract

For the longitudinal and lateral cooperative control problem of unmanned autonomous car (UAC) queue in complex environment, a multi-agent Deep Deterministic Policy Gradient (MADDPG) algorithm is proposed based on the UAC queue longitudinal and lateral cooperative control system in this paper. Firstly, in order to decouple the problem of longitudinal and lateral motion, MADDPG is used to control the longitudinal and lateral motion of the following cars separately, so that different agents can explore the environment space independently. Secondly, an action reward mechanism is introduced for each time step, which makes the system have faster response speed and higher stability when facing complex environments. Finally, comparative experiments are designed to show better security, efficiency and stability of the MADDPG-based UAC queue control system over multi-agent Deep Q-Network (MADQN).