Drones and autonomy in the sky

​Drones have gone from a hobbyist category to one of the most important frontiers of real world technological development. A large part of that comes from the battlefield; the war in Ukraine has turned cheap, mass produced drones into a defence strategy, and armed forces (and their budgets) everywhere have been scrambling to react ever since. This is not, however, solely a military story. Drones providing delivery services, infrastructure inspections, agricultural services such as crop spraying, and search and rescue are all viable commercial categories, with significant progress in recent years.

​Underneath, however, sits the same hard problems. Autonomy: can a drone fly beyond visual line of sight without a human approving every move? Particularly when it only has a the compute and power budget the size of a phone. And to what extent are we comfortable with autonomous decisions being made in this arena, with privacy concerns for those on the ground, and safety concerns for those already in the air?

​Much of the autonomy question comes down to navigation. GPS is easy to jam or spoof, so a drone increasingly has to work out where it is from cameras and motion sensors alone, with no satellite fix to fall back on. Coordinating more than one drone raises additional questions; a swarm has to make decisions on patchy, low-bandwidth communication rather than waiting on a single controller. Underneath there are physical constraints: battery energy density, which determines how long a drone can fly, and questions around computing choices; such as small GPU-style chips, custom low-power processors or other approaches.