Influence of Network Latency on Collective Perception Aggregation in V2X Networks

Autonomous vehicles and driver assistance systems depend heavily on the quality of their perception algorithms and sensors. Although the advances in hardware and software have increased the confidence in perception data, the field-of-view limitation cannot be overcome by single-vehicle perception. Cooperative Perception aims to solve the field-of-view limitations by employing perception data transmitted over vehicular networks. The vehicular network is still in its early stages of real-world implantation. Therefore, the effectiveness of cooperative perception relating to the market penetration of connected vehicles and infrastructure remains to be seen. In addition to the number of vehicles transmitting perception data, the drive to reach a Cooperative, Connected, and Automated Mobility paradigm must contend with the inherent lack of confidence in network data that cannot be blindly trusted. In this work, we propose a Collective Perception Aggregation Pipeline as a basis for collective perception data verification. The pipeline's inputs accept any data source that produces high-level perception data, such as sensors, perception algorithms, and cooperative perception messages. We investigate network latency's effects on the pipeline's data fusion component through simulated vehicular scenarios. Although the increase in message volume caused a higher channel busy ratio, the increased latency did not affect the prediction error of the data fusion algorithm. In a simulated scenario of 90 connected vehicles transmitting collective perception data, with a channel busy ratio of more than 20% and more than 50% of perception messages arriving out-of-sequence, the fusion algorithm maintained a standard deviation of the prediction error of 0.5609 m for the X position, 0.5627 m for the Y position, and 0.6267 m for the yaw of the object.