A Framework for Assessment of Perception Systems in Autonomous Vehicles

Sensors are integral components of autonomous vehicles and serve as vital interfaces that allow vehicles to perceive and interact with their surroundings. Although existing research has focused primarily on sensors in specific road settings, limited attention has been paid to a standardized methodology to understand their efficacy in urban and unconventional environments. This study focuses on common sensor types and their performance in urban terrains with an outlook for unconventional terrains, bridging a critical gap in the literature. A framework is proposed to quantify sensor performance in various environments, using experiments and simulations for data acquisition. The framework facilitates precise analysis, enabling researchers and developers to effectively evaluate sensor capabilities. To validate this framework, a simulated scenario was employed using Carla to observe the behavior of an autonomous vehicle. The results demonstrate that the sensors with the respective algorithms in the simulation are adapted to different lighting conditions, ensuring consistent sensory performance. The framework enables robust localization and kinematic estimations, enhancing the analysis of vehicle motion using state-of-the-art simulators. This research contributes to a deeper understanding of sensor performance and a validated framework to assess this performance, offering a valuable tool for advancing autonomous vehicle technology.