No abstract

Connected vehicles are expected to become key players in using information and communication technology in road transportation. They also hold great potential as a sensing platform. Since connected vehicles are equipped with various high-end sensors and communication modules and move around a city with virtually no risk of running out of battery, they are ideal as mobile sensor nodes. If camera images and LiDAR (light detection and ranging) * pointcloud data from tens of millions of connected vehicles could be collected, it would be possible to continuously scan an entire city and build a digital twin of that city. Assumed use caseUnfortunately, the total amount of sensor data generated by tens of millions of connected vehicles ranges from 10 to 100 Tbit/s. This is too enormous for communication networks, computers, and storage units to handle. This means that it is impractical to collect all available sensor data. It should be noted that sensor data include not only data that should be collected immediately but also data for which some collection delay is tolerable, data for which periodic collection is sufficient, and data that are of no value. Therefore, it is important to selectively collect important sensor data on a priority basis. By adjusting the pace and timing of sensor-data collection in accordance with the amount of load on the communication network and computers, it is also possible to equalize the load fluctuation over time, thus improve facilityutilization efficiency.The collaboration projects between Toyota Motor Corporation and the NTT Group include several use cases that require the selective collection of sensor data. This article focuses on the obstacle-detection use case and introduces the technical challenges, implementation details, and our efforts to improve the performance of selective collection. In this use case, obstacles on the road are assumed to be moved by the wind or removed by the road administrator. Therefore,