A distributed multi-vehicle architecture for autonomous driving simulation with an application to autonomous valet parking

Abstract

Simulation platforms are essential for developing and validating autonomous driving systems before real-world deployment. However, most existing simulators are designed for single-vehicle scenarios and provide limited support for distributed multi-vehicle experimentation across multiple computing hosts. This thesis presents the Distributed Multi- Autonomous Vehicle Architecture (DMAVA), which enables multiple AVs to operate concurrently in a shared simulation environment while maintaining independent autonomy stacks. The architecture integrates Autoware, AWSIM Labs, ROS 2 namespaces, and Zenoh-based cross-host communication to support distributed experimentation. Building on this framework, the Distributed Multi-Vehicle Autonomous Valet Parking System (DMV-AVP) is introduced as an application layer that integrates infrastructure-assisted perception through a YOLOv5-based overhead camera module and a multi-vehicle AVP coordination framework for managing shared parking resources. The system enables coordinated vehicle sequencing, exclusive parking spot allocation, and distributed vehicle state synchronization. Experimental validation using two-host and three-host deployments demonstrates stable localization, reliable inter-host communication, and coordinated multi-vehicle AVP behavior.