Distributed and Communication-Efficient Continuous Data Processing in Vehicular Cyber-Physical Systems

Licentiate thesis, 2020

Processing the data produced by modern connected vehicles is of increasing interest for vehicle manufacturers to gain knowledge and develop novel functions and applications for the future of mobility.
Connected vehicles form Vehicular Cyber-Physical Systems (VCPSs) that continuously sense increasingly large data volumes from high-bandwidth sensors such as LiDARs (an array of laser-based distance sensors that create a 3D map of the surroundings).
The straightforward attempt of gathering all raw data from a VCPS to a central location for analysis often fails due to limits imposed by the infrastructure on the communication and storage capacities. In this Licentiate thesis, I present the results from my research that investigates techniques aiming at reducing the data volumes that need to be transmitted from vehicles through online compression and adaptive selection of participating vehicles. As explained in this work, the key to reducing the communication volume is in pushing parts of the necessary processing onto the vehicles' on-board computers, thereby favorably leveraging the available distributed processing infrastructure in a VCPS.
The findings highlight that existing analysis workflows can be sped up significantly while reducing their data volume footprint and incurring only modest accuracy decreases. At the same time, the adaptive selection of vehicles for analyses proves to provide a sufficiently large subset of vehicles that have compliant data for further analyses, while balancing the time needed for selection and the induced computational load.