Resilience by design is mandatory to support the certification of the embedded/artificial intelligence of an autonomous swarm of drones

Several armies and private companies are considering swarms of drones [1], [2] as a mean to achieve missions on theatres of operations [3] because they offer several advantages. For instance, they make it possible to support continuous flight and they offer features that a single system cannot provide [4]. To be fully effective, a swarm should be autonomous, i.e. it should be able to take its own decisions in response to external or internal events, which requires some sort of embedded intelligence. Embedded intelligence is not artificial intelligence (AI) in itself, but it gives birth to a form of AI when autonomy is at play. A major characteristic (some would say drawback, but it is the prize to pay for the above advantages) of this kind of swarm, when considering a large number of aircraft, is that failure is the nominal mode of operation [5], because the more aircraft, the higher the probability of failure. The problem is that failure is by definition an obstacle towards certification, which is a
_sine
qua
non
condition to the wide deployment of these complex systems. It is therefore crucial to consider failure from the start of application design. We show below that distribution and resilience are key topics to address.