A Hybrid Scheduling Approach for Optimizing UAV-Assisted Edge Computing toward Efficient LLM Inference

Large Language Models (LLMs) play an important role in robotics and embodied intelligence system at the edge, enabling capabilities such as multi-modal data processing, real-time natural language interaction, complex task planning and autonomous decision making. However, numerous resource constrained devices cannot support the computationally intensive LLMs locally, which presents challenges in meeting quality-of-service requirements, end-to-end latency demands and bandwidth constraints for responsive operations. To address this problem, an UAV-assisted edge computing framework for LLM inference, incorporating a hybrid scheduling approach to optimize the LLM inference request scheduling and the energy consumption of the UAVs is proposed. A hybrid algorithm is proposed integrating estimation of distribution algorithm and evolution algorithms with high global exploration capabilities and efficient local exploration abilities. A series of simulation experiments is carried out to evaluate the performance of the proposed algorithm. The results demonstrate that the proposed algorithm outperforms existing algorithms in terms of average energy consumption, stability and code complexity.