Edge-Accelerated UAV Operations: A Case Study of Open Source Solutions

This study explores the execution of AI algorithms on open Unmanned Aerial Vehicles (UAVs) equipped with BeagleBone AI-64 (BBAI-64) boards, comparing their performance to high-performance computers equipped with GPUs. Key factors are evaluated, such as inference time, end-to-end processing time, CPU usage, or temperature on the board. Furthermore, this study presents the development of an open UAV platform based on an open-source flight controller (Durandal) executing an open-source autopilot (ArduPilot). This platform facilitates the integration of various sensors or cameras, regardless of brand or communication protocol. The study’s key findings show that the BBAI-64 offers advantages for smaller Artificial Intelligence (AI) models, and achieving comparable performance for larger models with high-performance computers. This work contributes to optimising AI execution on UAVs and supporting the development of versatile, sensor-agnostic open-source UAVs.