Green-EMulTO: A Next Generation Edge-Assisted Multi-Level Traffic Orchestrator for Green Computing in Consumer Autonomous Vehicles

Imaging technology in vehicular communications has advanced in the consumer industry. Owing to its ability of sensing ambient environment, decision control, and actuating the vehicle, the technology is able to enhance vision, improve traffic management, and offer convenience to the consumer autonomous vehicles. These vehicles in green computing environments require accurate obstacle detection (OD) and real-time video analytics to enhance on-road perception for forewarned accidents and pollution-free navigation. However, unforeseen obstacles in high vehicle speeds, adverse environments, etc., cause accidents leading to pollution. In this paper, we propose a Green-EMulTO, an <underline xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">E</u>dge-assisted <underline xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">Mul</u>tilevel <underline xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">T</u>raffic <underline xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">O</u>rchestrator that recognizes obstacles in low latency. It employs a priority queue for the traffic imaging streams, a bandwidth manager for V2X services, and a lightweight DNN model for fast on-device OD. We also introduce a Synergistic service placement and cost minimization algorithm (SSPCM) based on Lyapunov optimization and Markov approximation. It reduces the response latency by addressing the intrusive dynamics of video and unknown network fluctuations in the autonomous driving environment. This orchestrator is designed to provide a pollution-free environment by reducing road accidents thereby satisfying the green vehicular environmental goals. In addition, we develop an autonomous driving platform using NVIDIA JetRacer AI Pro Kit and Jetson Nano for system-level verification. We have also compared it with the benchmark lightweight models (YOLOV5-nano, YOLOv6-nano, YOLOV7-tiny, YOLOv5-small, YOLOv6-small) on such commercial devices. Green-EMulTO witnessed an improvement of up to 20% in accuracy and the training time was reduced to less than 60%. Hence, this orchestrator improved the real-time inference speed over different autonomous driving environments.