Cooperative Generative AI for UAV-Based Scenarios: An Intelligent Cooperative Framework
Longyu Zhou, Wenjiao Feng, Zihan Chen, Tianchen Ruan, Supeng Leng, Howard H. Yang, Yaru Fu, Tony Q. S. Quek
Abstract
The Internet of Things (IoT) has been expediting unmanned aerial vehicle (UAV) network autonomy for 5G and beyond applications. In this context, generative artificial intelligence (GAI) is widely applied to improve service efficiency. It can provide customized service decisions for diverse UAV requirements through deep learning. However, existing GAI solutions may fail to generalize well for mission-critical UAV scenarios due to dynamic changes in physical environments and limited computing resources. We propose a cooperative GAI framework to achieve a highly accurate and low-latency GAI implementation performance with a double-sized GAI manner. We first propose a model aggregation and splitting algorithm to acquire accurate large-sized GAI models by combining small-sized GAI models using a multimodal-based deep learning method. Then, we propose a model update algorithm to reduce the latency of GAI implementation for real-time service provisions. We demonstrate the effectiveness of our terminal-edge cooperative GAI framework using a case study of UAV-based target tracking. The results indicate that our solution ensures an accurate GAI implementation with a successful tracking ratio of up to 90% as well as a low system latency of under 2 s compared to the existing GAI pattern on average, respectively.