Co-Vast: A Simulation Platform for Vehicle Collaboration in Autonomous Driving Scenarios

In order to solve the problem of blind spots caused by limited sensory data of a single vehicle, multi-vehicle collaboration has been attracting more and more attention in the field of autonomous driving in recent years. To facilitate the study on multi-vehicle collaboration, we need construct realistic traffic scenarios and deploy a number of vehicles. If such scenarios are constructed in reality, the cost would be very high. Thereby, current studies on vehicle collaboration mainly rely on simulation platforms. However, existing simulation platforms are mainly designed from the perspective of a single vehicle, which provide quite limited support for the establishment of communication and collaboration among different vehicles.To fill this gap, we design and implement a distributed simulation platform (Co-Vast), aiming to enable studies on vehicle collaboration in autonomous driving scenarios. In particular, the proposed platform consists of three different types of components, which are field, vehicle agent, and infrastructure agent. The field supports the simulation of real-world scenarios, in which both vehicle and infrastructure agents can be created and managed dynamically. Moreover, a communication proxy is designed to support efficient inter-vehicle communication. To validate the performance of the platform, we quantify its resource consumption from different perspectives. Results show that the CPU usage of Co-Vast does not exceed 2%, and the communication delay is around 25ms, indicating Co-Vast can be used to facilitate the simulation of large-scale vehicle collaboration.