Enhancing Autonomous Vehicle Safety and Efficiency: Development of a Scaled Testbed for Evaluating Multiple ANN Architectures in Autonomous Driving.

Abstract

Modern transportation is facing increasing challenges, such as high rates of accidents due to human error, insufficient reaction time, and decision efficiency of humans, which require improvement in advanced autonomous vehicle (AV) technologies. On this basis, this research aims to develop a small-scale testbed for experimental driving via Artificial Neural Networks (ANN) and Computer Vision, more specifically Convolutional Neural Networks (CNN). Key AV tasks like lane detection, road sign recognition, steering control, and obstacle detection can be controlled during experimentation on a testbed. The processing of data from an on-board camera from the CNN models is used for real-time perception and two decisions, unlike the reinforcement learning-based approaches. Thus, this study objectively evaluates the performance of the ANN models under various driving conditions concerning response time and navigation accuracy. It is shown further how CNN-based techniques can improve the perception and control of AV. This work addresses the cost-effective and scalable platform that bridges the gap between simulation-based research and real-world AV experimentation compared to existing studies. The importance of the findings is in developing safer, more reliable autonomous vehicles and likewise providing a practical framework for further developments in AV technology.