Multi-Metric evaluation of Attention-Based Neural Network Architectures for Autonomous Driving

Abstract: Autonomous driving is the capability of a vehicle to drive on its own. Self-driving cars can reduce human efforts, transportation cost, number of deaths in road accidents. Deep Learning (subset of Neural Network) has huge number of models with which the research on autonomous driving is made easy. Attention is a mechanism that gives different type of results when it is integrated with different Deep learning models. This paper evaluates some important Attention Mechanisms of Transformer model by calculating multiple metrics for each Mechanism.

Keywords: Collision rate, Driving Score, F1 Score, IoU, mAP, Route Completion.