• Acceleration
• Drones
• FPGA
• HLS
• KCF

Small Unmanned Aerial Vehicles (SUAV), also called drones, are cyber-physical systems that typically rely on complex computer vision algorithms or even neural-based perception functionality with real-time requirements. For this reason, high-performance embedded platforms with support for hardware acceleration are commonly adopted to accomplish the mission. However, SUAVs have limited energy budgets (they are typically powered by batteries), and so power consumption and weight shall be kept as low as possible. This work aims at realizing a new, energy-efficient computing platform for the Pitom drone, a vehicle provided by the AITronik company initially powered by a GPGPU-based platform. This work proved that alternative FPGA-based platforms can bring significant improvements in terms of weight, power consumption, and predictability while maintaining the same performance. The on-board software is powered by CLARE, a hypervisor-centric software stack that provides a safe, secure and predictable environment for mixed-criticality applications. In particular, the system is composed of two domains: (i) Linux, which is responsible for acquiring images from the camera and performing the tracking algorithm; and (ii) Erika Enterprise 3, a hard Real-Time Operating System (RTOS), which is responsible for piloting and controlling the vehicle. Finally, the system has been tested in hardware-in-the-loop by using Unreal Engine and AirSim to simulate both the external environment and the drone