Serge Chaumette, Professor at the University of Bordeaux/LaBRI (Laboratoire Bordelais de Recherche en Informatique, UMR CNRS 5800), head of the ASPIC (Autonomous System, Perception Interaction & Control) Master's program and head of drone activities at LaBRI, he is also co-founder and Innovation Director of the start-up Preditic, which works in the field of (predictive) industrial maintenance using sensors, proximity communication technologies and AI where it makes sense. He is also the scientific director of IcarusSwarms.ai, which is dedicated to the evaluation of anti-drone systems and military applications of swarms with products already used by the armed forces.
He shares his expertise on the impact of artificial intelligence in the field of drones, exploring how this technology is transforming the future of autonomous flight and mission precision.
AI and drones
First of all, when we talk about AI and drones, we must first separate the two technologies in order to better address them holistically. AI itself brings a number of unique capabilities, and drones also offer unique potential. It is the combination of these two technologies (and many others) that has led to the emergence of the functional power of drones as we have seen it in recent years.
How does artificial intelligence improve the autonomy and decision-making capabilities of drones, in both civil and military applications?
AI can be used at several levels of the technological stack that makes up a drone system. It can help process data on board the device directly but also on the ground, for example in a C2 for the military domain. Typically, it involves the analysis of data capture (conventional camera, thermal camera, chemical nose, etc.) to help with situation management and decision-making. These decisions can ultimately be made by/in the device itself, which gives it a certain degree of autonomy, or by operators/decision-makers.
AI is thus a catalyst for the data captured in the drone's environment. It offers an analytical capacity that eliminates the need for human intervention at certain levels of the preparatory work for decision-making, thus giving the device a certain degree of autonomy. However, we must remain cautious. Indeed, when we talk about autonomy, we often think of decision-making without any human intervention (and we immediately think of killer robots, killer drones). Clearly, we are not at that stage, both for technical and ethical reasons.
What are the main limitations and precautions to be taken in the use of AI to avoid abuses, particularly in terms of security and ethics?
AI does not offer a guarantee of success. The results are often probabilistic and therefore some results from an AI engine may be wrong. To be very concrete, an image processing AI, for example, will be able to identify a civilian device as an opposing military device in the absence of suitable learning data. We are talking about hallucination here and we clearly understand the risk of leaving the choice of a potentially lethal decision to the system.
This ties in with the issue of the explainability of AI. To continue with our example, AI will be able to differentiate between types of vehicles by classifying them into different categories, but the reason for classifying an image in one category or another is not explained. The trust that can be placed in it is therefore a key issue.
This is why the approach of keeping a man in the loop (man in the loop) remains fundamental and is the one adopted by most countries. A decision, if it is impactful, must be that of the man and not of the machine.
In the field of drones (in particular but not exclusively), the issue of learning, which is the basis of the quality of the resulting system, is therefore a delicate subject. This learning is based on the analysis of a very large quantity of data sets. However, these significantly sized data sets do not always, or not yet, exist in the field of drones.
Finally, while the ability to remotely pilot a drone brought the decision-maker and the experts closer to the theater of operations, autonomy, in a way, distances them from it, which can be seen as a risk of disempowerment by offloading the decision to the machine.
Can you give us any concrete examples of projects or research where AI has enabled significant advances in the field of drones?
Today, significant advances are being made in the field of drones themselves and the missions they can accomplish. While AI can help in the design of devices, large or small, it is naturally the missions that it makes possible that are the primary focus. This takes the form of embedded technological building blocks (navigation management, obstacle avoidance, intelligent energy management, etc.) and increasingly varied and complex missions: mine detection, search and rescue, sewer exploration, equipment transportation, etc.
Does AI make it possible to envisage a more fluid collaboration between drones and human operators? What are the challenges to be met in order to optimize this interaction?
AI makes it possible to relieve the operator of a certain number of procedures. It also makes it possible to eliminate a large amount of non-significant information. For example, in the context of surveillance, we talk about false positives. What is new is, thanks to the combination of drone and on-board AI, the ability to do this during an operation without requiring interaction with the ground and without human intervention. This means that the information that is fed back is more relevant, less numerous, and the mental load on the operators is reduced.
In your opinion, what will be the major trends and innovations of tomorrow in terms of intelligent drones, and what role can France play in this development?
Swarms are a fundamental issue. We are seeing this today in Ukraine and recent trade shows clearly echo this. These configurations provide additional capabilities (it's 1+1 better than 2) by combining devices of different sizes, elongation, autonomy and sensors. Moreover, being able to dynamically adapt the composition of a swarm to meet the needs of an operation is a definite plus. Heterogeneous fleets combining aerial drones, ground robots and surface drones are thus one of the challenges of the future.
Tactical deployment capability is also a real need. Adapting as closely and as quickly as possible to the problems on the ground is an operational necessity. Thus the market for small and medium-sized drones (light, inexpensive, stealth, etc.) has developed, even if, of course, HALE and MALE drones make perfect sense, particularly in the military field.
The multiplicity of payloads is also fundamental. The days when devices constituted a differentiator are over, as we have seen through the consolidation of the market. Moreover, drone manufacturers no longer go to drone shows where drone specialists talk to drone specialists; they go to professional shows (agriculture, surveillance, military, etc.). The UAV show stands out for its scale and expertise, its broad spectrum and its location in a pioneering territory in the field where the drone is part of a strong technological tradition and continues to develop, both in the military and civilian fields.
Clearly, the increase in the processing capacity of on-board processors will continue to favor the development of new systems (in the field of AI but not only).
To conclude, it can be said that the drone is at the crossroads of technological innovations, not only of AI, it is the result of the convergence of scientific and technical progress. We could thus talk about quantum for sensors, navigation and communication security, batteries and energy in general, etc.
