Towards a Drones-as-a-Service Platform for Application Programming

2025-04-04

AI Summary

Deploying deep learning applications on unmanned aerial vehicles (UAVs) faces challenges stemming from hardware heterogeneity, stringent real-time requirements, and complexities in cloud–edge collaboration. Method: This paper proposes a service-oriented “Drone-as-a-Service” (DaaS) architecture that abstracts underlying hardware diversity and provides composable microservice primitives for on-demand perception, autonomous navigation, and collaborative edge–cloud analytics. It introduces a hardware-agnostic, cross-platform intelligent decision-making framework integrated with a lightweight service orchestration mechanism. Contribution/Results: The architecture achieves ≤20 ms per-frame latency and ≤0.5 GB memory footprint. Evaluated on embedded AI platforms—NVIDIA Jetson Orin Nano and DJI Tello—within an AWS cloud environment, it supports four representative application scenarios. Application development requires only ~40 lines of code, significantly reducing implementation complexity while enhancing deployment flexibility and system scalability.

Technology Category

• Machine Learning: Hardware-aware ML
• Computer Vision: Learning & Optimization for CV
• Intelligent Robots: Multi-Robot Systems

Application Category

• Systems and Infrastructure for Web, Mobile and WoT: Applied ML and AI for Web-based mobile applications
• Responsible Web: Human-perceived consequences of algorithmic deployment on the web
• Security and Privacy: Security and privacy of machine learning and AI applications

Abstract

The increasing adoption of UAVs with advanced sensors and GPU-accelerated edge computing has enabled real-time AI-driven applications in fields such as precision agriculture, wildfire monitoring, and environmental conservation. However, integrating deep learning on UAVs remains challenging due to platform heterogeneity, real-time constraints, and the need for seamless cloud-edge coordination. To address these challenges, we introduce a service-oriented framework that abstracts UAV-based sensing complexities and provides a Drone-as-a-Service (DaaS) model for intelligent decision-making. The framework offers modular service primitives for on-demand UAV sensing, navigation, and analytics as composable microservices, ensuring cross-platform compatibility and scalability across heterogeneous UAV and edge-cloud infrastructures. We evaluate our framework by implementing four real-world DaaS applications. Two are executed using its runtime on NVIDIA Jetson Orin Nano and DJI Tello drones in real-world scenarios and the other two in simulation, with analytics running on edge accelerators and AWS cloud. We achieve a minimal service overhead of<=20 ms per frame and<=0.5 GB memory usage on Orin Nano. Additionally, it significantly reduces development effort, requiring as few as 40 lines of code while maintaining hardware agnosticism. These results establish our work as an efficient, flexible, and scalable UAV intelligence framework, unlocking new possibilities for autonomous aerial analytics.

Problem

• Integrating deep learning on UAVs with platform heterogeneity
• Ensuring real-time AI-driven applications with cloud-edge coordination
• Reducing development effort for UAV-based sensing and analytics

Innovation

• Service-oriented framework for UAV intelligence
• Modular microservices for sensing and analytics
• Minimal overhead with cross-platform compatibility