Collaborative Perception and Computing Offloading in 6G Air-Ground Integrated Networks
Hongbo Jiang, J. Guo, Zhu Xiao, Kehua Yang, Tong Li, Bo Li
Abstract
The evolution of sixth-generation (6G) wireless communication significantly accelerates the Internet of vehicles innovation, catalyzing advancements in autonomous driving systems. The collaborative model utilizing 6G is expected to break through the vehicle’s inherent field-of-view deficiencies and heterogeneous computational resource constraints, further improving the efficiency of the technology. This paper proposes a novel 6G NOMA air-ground integrated sensing-computing framework that achieves high-quality collaborative perception and low-latency 3D computing offloading to alleviate restrictions through cooperative networking with unmanned aerial vehicles (UAVs) and road-side units (RSUs). To balance latency and UAV energy consumption for efficient collaboration, we formulate it as a mixed integer nonlinear programming problem (MINLP). Considering the time sensitivity of the perceptual task, we introduce queuing and Lyapunov optimization theory to transform the optimization objective into a Lyapunov drift-penalty function and derive its upper bound, which we model as a Markov decision process (MDP) and optimize it with Large Language Models (LLMs) assisted temporal replay deep reinforcement learning (TR-DRL). For perception quality improvement, an elite-guided binary-weighted firefly algorithm is developed to solve combinatorial optimization in perception fusion. Experimental results demonstrate 4.88% and 8.26% improvements in latency and energy efficiency respectively, alongside enhanced perception fusion quality 7.41% compared with advanced counterparts.