A New Vehicular Fog Computing Architecture for Cooperative Sensing of Autonomous Driving

The sensing coverage and accuracy of vehicles are vital for autonomous driving. However, the current sensing capability of a single autonomous vehicle is quite limited in the complicated road traffic environment, which leads to many sensing dead zones or frequent misdetection. In this paper, we propose to develop a Vehicular Fog Computing (VFC) architecture to implement cooperative sensing among multiple adjacent vehicles driving in the form of a platoon. Based on our VFC architecture greedy and Support Vector Machine (SVM) algorithms are adopted respectively to enhance the sensing coverage and accuracy in the platoon. Furthermore, the distributed deep learning is processed for trajectory prediction by applying the Light Gated Recurrent Unit (Li-GRU) neural network algorithm. Simulation results based on real-world traffic datasets indicate the sensing coverage and accuracy by the proposed algorithms can be significantly improved with low computational complexity.