90 Focus | Autopilots if they have four autopilot computers and one GNSS-IMU. Hence, complete autopilot solutions designed for higherrisk, higher-end applications are being built with not just multiple microcontrollers but growing numbers of GNSS receivers and antenna inputs, IMUs, data-link interfaces, and traditional aviation sensors such as pitot tubes and barometers, which have been proven to be reliable sources of information over their decades of use. While all these systems are key to a good autopilot system, some bear closer scrutiny than others. For instance, a good gyroscope defines for many the quality of the whole autopilot, as having reliable angular rate data makes countless potential autopilot faults easier to resolve. But, for sheer controllability, choosing a central processing unit (CPU) with the right architecture and internal features is the key to safeguarding against dire issues, such as encryptions against the hostile seizure of data from an uncrewed system, advanced manoeuvres needed for flying amid obstacles or below radar, or smart behaviours for flight safety amid jamming or spoofing (which we will discuss in more detail). It is increasingly common for integrators to opt for a GPU or a system-on-chip (SoC) type of machine to handle such advanced functions – and a range of robust systems are available – but they take up significantly more weight and power than a CPU, and are far easier for hostile actors to detect, given their thermal and electromagnetic (EM) emissions. Hence, some UAV autopilot manufacturers remain committed to choosing the right CPU for integrating precisely the code needed for the mission and environment, and condensed to avoid wasting weight, computational power and electrical power, especially if aviation laws end up requiring triple-redundant CPUs for autopilots to be certifiable. Naturally, the gradual increase in redundancy requirements comes with certain risks to uncrewed systems engineering and manufacturing. For one, making an autopilot triple-redundant will obviously triple the number of chips it needs, driving up the pressure on already strained supply chains of silicon. A more redundant autopilot inevitably requires more core and ancillary systems to support it, driving up the size of the UAV required to carry a given payload mass. Hence, autopilot manufacturers are carefully scaling and selecting parts with SWAP to reduce the need for a 40 kg UAV to deliver an 800 g payload. Autopilot developers are working on alternate configurations and means for guaranteeing that uncrewed systems continue operating normally, keeping within safe boundaries in the event of key subsystems failing. GNSS-independent configurations A few new systems are coming with integrated or optional vision systems for smart use of optical positioning references to compensate for drops in GNSS or IMUs, and to satisfy authorities that loss or spoofing of GNSS-inertial inputs will not result in losing the aircraft or its navigation integrity. This is critical as even triple-redundant, multi-constellation GNSS receivers may be undone by a singular jamming attack. Additional observer functions and sensor integrations for position estimation will be increasingly critical to certifiable autopilot arrangements in the future. Visual navigation is now widely used by indoor and urban UGVs, given that GNSS struggles in cities and buildings, but some UAV autopilots are now integrating cameras and machine-vision systems for recognising specific points or patterns of terrain amid GNSS spoofing or jamming. It follows that extremely robust logic for detecting spoofing – such as constant cross-referencing of GNSS data with accelerometer or pitot tube readings to immediately notice conflicting speed data indicative of GNSS inaccuracies – is imperative for autopilots to determine when to disregard GNSS data. For effective visual navigation capabilities, it is prudent that autopilots can either be embedded with, or generate in real time, their own georeferenced imagery of the ground to leverage terrain visuals in known areas for accurate self-geolocation. In unknown areas, self-geolocation may be impossible with vision alone, but a robust visual odometry can minimise drift to within 1% error, and potentially enable a UAV to return to a known area or its launch and recovery point. Some autopilots are being designed to leverage inputs other than visions of terrain and ground structures, as thick cloud cover beneath high-altitude UAVs or the absence of terrain for USVs and maritime UAVs would leave such April/May 2024 | Uncrewed Systems Technology Using an autopilot with three or more cores is one way to assure the redundancy and airworthiness of a professional UAV (Image courtesy of Embention)
RkJQdWJsaXNoZXIy MjI2Mzk4