[Paper Review] Distributed Machine Learning for UAV Swarms: Computing, Sensing, and Semantics

Yahao Ding, Zhaohui Yang+3 more|arXiv (Cornell University)|Jan 3, 2023

UAV Applications and Optimization4 citations

TL;DR

This survey presents a comprehensive analysis of distributed machine learning (DL) techniques—federated learning, multi-agent reinforcement learning, split learning, and distributed inference—for unmanned aerial vehicle (UAV) swarms. It demonstrates how these DL methods enable privacy-preserving, scalable, and adaptive coordination in UAV swarms for applications in wireless communication, remote sensing, VR delivery, and semantic communications, with key results showing improved convergence, reduced latency, and enhanced task completion rates in dynamic environments.

ABSTRACT

Unmanned aerial vehicle (UAV) swarms are considered as a promising technique for next-generation communication networks due to their flexibility, mobility, low cost, and the ability to collaboratively and autonomously provide services. Distributed learning (DL) enables UAV swarms to intelligently provide communication services, multi-directional remote surveillance, and target tracking. In this survey, we first introduce several popular DL algorithms such as federated learning (FL), multi-agent Reinforcement Learning (MARL), distributed inference, and split learning, and present a comprehensive overview of their applications for UAV swarms, such as trajectory design, power control, wireless resource allocation, user assignment, perception, and satellite communications. Then, we present several state-of-the-art applications of UAV swarms in wireless communication systems, such us reconfigurable intelligent surface (RIS), virtual reality (VR), semantic communications, and discuss the problems and challenges that DL-enabled UAV swarms can solve in these applications. Finally, we describe open problems of using DL in UAV swarms and future research directions of DL enabled UAV swarms. In summary, this survey provides a comprehensive survey of various DL applications for UAV swarms in extensive scenarios.

Motivation & Objective

• To address the challenges of centralized machine learning in UAV swarms, including limited energy, computational capacity, and data privacy.
• To investigate how distributed learning enables collaborative, real-time decision-making in dynamic, mobile UAV swarm environments.
• To analyze the integration of DL with emerging technologies like RIS, VR, and semantic communications for next-generation UAV networks.
• To identify open research problems and future directions in DL-enabled UAV swarm architectures, communication efficiency, and real-world testbed development.

Proposed method

• Categorizes and compares four core DL paradigms: federated learning (FL), multi-agent reinforcement learning (MARL), split learning, and distributed inference.
• Analyzes the application of these DL techniques to UAV swarm tasks such as trajectory design, power control, resource allocation, and user assignment.
• Examines the use of Markov decision processes (MDPs) and deep reinforcement learning (DRL) frameworks like TD3 for optimizing UAV trajectory and VR rendering modes.
• Proposes joint optimization of UAV location, communication, computing, and caching policies using iterative algorithms to solve non-convex problems.
• Introduces over-the-air computation and semantic communication frameworks to reduce wireless transmission overhead and improve efficiency in swarm communication.
• Proposes a hybrid offline training and online running approach using DRL and game theory for scalable and adaptive VR content delivery.

Experimental results

Research questions

• RQ1 How can federated learning enable privacy-preserving model training in UAV swarms without sharing raw sensor data?
• RQ2 What role does multi-agent reinforcement learning play in enabling autonomous, dynamic decision-making for UAV swarm coordination?
• RQ3 How can split learning and distributed inference reduce computational load and communication overhead in large-scale UAV swarms?
• RQ4 In what ways can distributed learning enhance performance in emerging UAV applications such as VR delivery and reconfigurable intelligent surfaces (RIS)?
• RQ5 What are the key architectural, communication, and computational challenges in deploying DL for real-world UAV swarm systems?

Key findings

• The use of DRL with the TD3 algorithm significantly improved convergence speed and rendering completion rate compared to baseline DDPG in VR rendering tasks.
• Joint optimization of UAV trajectory, caching policy, and computing resource allocation achieved superior performance in reducing backhaul latency for VR content delivery.
• Distributed inference and split learning enabled effective training of large ML models across UAV swarms by partitioning models and exchanging only intermediate features.
• Semantic communication frameworks were shown to support task-oriented, efficient wireless transmission among UAVs, reducing redundant data exchange.
• Over-the-air computation techniques were identified as effective for handling massive access and low-latency requirements in large-scale UAV swarm networks.
• Real-world testbeds and experimental validation are critical for verifying the scalability and adaptability of DL-based UAV swarm systems in dynamic, real-time environments.