Sensor fusion for autonomous vehicle environmental perception: a reproducible comparison

Abstract

Autonomous vehicles (AVs) rely heavily on advanced sensor technologies to ensure safe navigation and obstacle avoidance in dynamic and diverse environments, yet single sensors have inherent limitations in perceptual information completeness, environmental adaptability and functional safety, failing to support stable autonomous driving in complex open scenarios. Sensor fusion thus becomes a core approach for autonomous driving perception systems, which synthesizes multi-source sensory data to boost perception accuracy, enrich environmental cognition and enhance system robustness beyond individual sensor capabilities. This paper reproduces and compares two different sensor fusion methods, intending to offer practical references for the real-world application of such systems in autonomous driving.