Offloading autonomous vehicle machine learning algorithms to the 5G edge: A proof of concept implementation

This paper investigates the effect of offloading machine learning autonomous vehicle algorithms to the edge of 5G networks. It describes the implementation of a field test experiment on the Canadian ENCQOR 5G network using edge computing to command a 1/10
^th
scaled RC vehicle. The vehicle uses a camera feed to predict steering. Predictions are computed on a server on the edge of the network which contains a pre-trained neural network. It was found that the autonomous vehicle performance are very similar with and without offloading. The paper discusses the limits of this result and provides an overview of future opportunities and obstacles in the use of offloading machine learning functionalities for autonomous vehicles in 5G.