Publication

Abstract : The ever-changing transportation industry will only get better with autonomous vehicles (AVs) that will make transportation safe, efficient, and intelligent. Even after considerable achievements, complex issues persist for achieving robust perception, real-time decision making, and scalable system architecture. This study suggests a modularized fusion-based AV where we connect solid-state LiDAR, monocular cameras, and 9-DOF IMU, with the backend stage being a Visual-Inertial-LiDAR SLAM and the perception layer being a Visual Transformer (ViT). Real-time cooperative perception by the 5G NR-V2X enables low-latency actuation by a hybrid PID-MPC control system, accelerated on FPGAs. The hybrid system overcomes shortcomings in latency and flexibility present in traditional architectures (Tesla, Waymo, and Apollo Auto). Comparative studies show advancements in end-to-end latency (∼60 ms), cost effectiveness, and semantic comprehension. Its modularity and edge-native design explain why the architecture can be used in all kinds of cities. This paper will help develop AVs that are resilient under future environments and connected to the infrastructure plans of smart cities.