Edge technologies in the automotive industry: An experimental latency evaluation of AWS Greengrass usage

Abstract

The digital transformation of the automotive industry is one of the most significant changes to the industry in its 140-year history. This transformation bring new functionalities to vehicles and changes how a vehicle is being used. In conjunction with the automotive industry’s digital transformation there has been a trend to move all computational power to the cloud. The cloud offers an efficient way to deal with data, but the network has started to become a bottleneck. The purpose of the study is to evaluate how the usage of edge technology can reduce latency and bandwidth usage in the automotive industry. The study is based on a controlled experiment where a prototype system was developed to evaluate different implementations of edge technologies. The system consists of an accelerometer connected to a car, and a machine learning algorithm that uses the data from the accelerometer to try to predict the type of road surface the car is driving on. Three concrete experimental setups was evaluated: One system where all processing is done in the cloud. One system where the pre-processing is done in an edge computational layer inside the vehicle instead of the cloud. And finally, one system where all processing is performed in an onboard edge computational layer. When testing the different implementations different bandwidth limitations were enforced as well. The study showed that moving all computational power to an edge node directly in the car reduces latency compared to other implementations, no matter the bandwidth limitation. The study also showed that with high bandwidth available, a system only running in the cloud could be faster than a system that uses both edge and cloud computing. However, when having a limit on the available bandwidth, a system that uses both edge and cloud computing is faster than a system only running in the cloud.