Guidelines for reducing perception errors in autonomous vehicle system architecture

Abstract

Autonomous vehicles (AVs) are poised to reshape the future of transportation, and the autonomous driving system impacts the overall safety of the driving system. Effective operation of AVs requires a vast deployment of sensor technologies for environmental perception and to acquire information on road conditions. While reliable perception is crucial for the success of AVs, perception errors represent a critical technical challenge in Avs. They can result in incorrect decision-making and unsafe driving behaviours. These perception errors significantly compromise vehicle safety and impede the successful commercialisation of autonomous driving systems. This study employed the Design Science Research Methodology (DSRM) process to develop a set of guidelines to minimise the perception errors of AVs, leading to a safer and more reliable autonomous driving experience. The development of these guidelines was a consequence of the triangulation of extant literature, internal validation and empirical data gathered through expert reviews using an online questionnaire. The study followed an embedded mixed-methods approach. A wide group of experts was drawn from the automotive industry in South Africa, the United States of America (USA), Hungary, and the United Kingdom (UK). Some of the experts were sourced through a reputable online platform, Prolific. The descriptive data analysis method was employed to analyse and interpret the data. Seven perception layer tasks discovered from the literature were presented to experts during the evaluation process. The results validated the importance of the seven perception layer tasks (viz., Object Detection and Classification, Lane Detection and Tracking, Real-Time Feedback, Assess Potential Obstacles, Sensor Data Fusion, Dynamic Adaptation, Localisation), and four additional tasks emerged from the evaluation process (viz., Motion Forecasting, Safety System Integration, Free Space Detection, and Weather Condition Detection and Response). The main contribution of this study is a set of guidelines, deemed potentially effective for mitigating perception errors in AVs. Importantly, the results validated the relevance of the preliminary guidelines, improved the quality of their design and underscored the need for further investigation into the implementation of their requirements.