Toward Inference Latency Optimization for Scalable Collaborative Multi-UAV Analytics

Ying Wang

doi:10.1109/TGCN.2025.3625726

Toward Inference Latency Optimization for Scalable Collaborative Multi-UAV Analytics

2025-12-23

0

PRE

AI

DOI：10.1109/TGCN.2025.3625726

Original

Original request for help

Save

Abstract

En 中文

Collaborative multiple uncrewed aerial vehicles (UAVs) demonstrate significant potential for real-time video analytics applications. Current multi-UAV systems face challenges such as inference latency and endurance. These problems primarily stem from limited computational resource and energy constraints of UAVs. The scale of UAV deployment is a crucial factor, as it imposes varying degrees of limitations on both inference latency and UAV endurance. This paper proposes a scalable cooperative UAV architecture for video analytics, which is optimized for different UAV scales and suitable for both centralized and distributed control modes. To minimize inference latency and enhance energy efficiency, we develop mathematical models and optimization algorithms for UAV collaboration-based video analytics, addressing both centralized and distributed scenarios. The centralized method uses a two-layer optimization algorithm to jointly optimize UAV deployment and task scheduling (JDTSO), while the distributed method integrates multi-agent proximal policy optimization (MAPPO) with a directed acyclic graph (DAG) partition strategy (MAPDP). Extensive analysis and numerical results demonstrate the superior performance of the proposed architecture.

Keywords:

Collaborative multi-UAV

computation offloading

DAG partition

Toward Inference Latency Optimization for Scalable Collaborative Multi-UAV Analytics

Abstract

Journal

Organization