MPSoC4Drones: An Open Framework for ROS2, PX4, and FPGA Integration

• Main Authors: Nyboe, Frederik Falk, Malle, Nicolaj Haarhoj, Ebeid, Emad
• Format: Conference Proceeding
• Language: English
• Subjects: Field programmable gate arrays; Middleware; Object detection; Power demand; Throughput; Visualization
• ISSN: 2575-7296
• DOI: 10.1109/ICUAS54217.2022.9836055

Autonomous drones are facing a tough efficiency challenge due to limitations on the utilized processing hardware units. Among these limitations is the tradeoff between fast computing and low power consumption; between functional complexity and flight time. Recent progressions point to FPGAs for accelerating heavy processing. In this work, we present the MPSoC4Drones Framework; a novel framework for organizing FPGA-design and OS build projects. The framework combines tools for creating bootable images for the Ultra96-V2 board.We show how MPSoC4Drones organizes the build, combining the latest well-known tools for research and industrial drone development, Ubuntu 20.04, PX4 autopilot, and ROS2 middleware. We validate the framework through a computationally intensive deep learning use case implemented in the MPSoC4Drones framework. We show the superior throughput and low power consumption of FPGA processing compared to classical CPU and GPU approaches. Finally, we offer the full framework as open-source.