Aller au contenu principal

Safer, More Reliable and More Agile Drones

Research Article published on 14 January 2022 , Updated on 14 January 2022

(This article was originally published in L'Édition No.17)

When used alone or several at once, professional civilian drones have demonstrated a largely underused potential. With drone use being in full flight, researchers are looking at how they may improve the reliability, security, and adaptability of these machines. 

You can often hear it buzzing long before you can see it. Whether it has fixed or rotating wings, drones of all sizes have seen a rapid rise over the past few years. These Unmanned Aerial Vehicles (UAV), which can be controlled using a radio control or a smartphone, and were initially for military use, have now taken over civil society. With sensors and HD cameras, drones have flooded the hobbyist market. At the same time, their use in professional contexts has grown, whether it is to assist people, deliver parcels, watch over fields, or even inspect facilities. This expansion, however, is still limited. “Although there has been an explosion of intentions to use them, the allocation of flight licences by the French Civil Aviation Authority (DGAC) remains very slow,” explains Reda Nouacer, from the Integration of systems and technology laboratory (LIST – Université Paris-Saclay, CEA). The security aspect is also a factor. 

This licence is the much-wanted ‘golden ticket’ for operators. Without it, their mission cannot take place. In order to receive it, the applicant must first declare and detail the planned flight conditions (points of departure and arrival, schedule, the context in which the drone will fly, statistics-based meteorological conditions, etc.). “These aspects are also used when choosing the drone, the sensors, the required battery life, etc. Yet, the licence, once issued, is only valid for that specific mission. An operator cannot really buy as many drones as they have missions to carry out,” says Reda Nouacer. The challenge is therefore to adapt the mechanics of a drone to a new mission quickly and affordably, without calling into question the reliability of its certification or slowing down the process of obtaining a flight licence. 

 

When Modulation Becomes Easy

Within the COMP4DRONES project, Reda Nouacer and his colleagues are working to develop an embedded architecture solution and a core set of reusable and enabling tools and technologies to be made available to SMEs to help them rapidly adapt their drones. Fortyeight academic and industrial partners are involved in this European Union funded project (ECSEL JU). 

They have taken inspiration from the modularity used in the automotive and avionics industries to develop agile customisation. “We are reusing an already-authorised module. The modular architecture that we have developed does not change the structure nor the design of the system. This ensures that the allocated licence remains valid and saves time.” They have identified a reference architecture, which means uniform and standardised interfaces, protocols, and integration methodology. In particular, they compared open-source autopilots (the flight software that is generally provided by the manufacturer) and identified their similarities. “We identified function-related building blocks, typical composition and the architecture patten.” In other words, the connecting mechanics between blocks. “It’s the equivalent of an integration and development protocol.” 

This type of suggested adaptation will make life easier for SMEs. “We can artificially increase their skills by providing a tool that will help them to develop the electronics and software that is to be integrated.” For drones, these elements are particularly important for their real-time detection of the environment. As their payload is limited, not everything can be embedded straightaway. “This is what we’re adding to the system to allow it to make safe and autonomous decisions. The tools that we are proposing have the added benefit of having already been tested.” Tools such as Papyrus modeler, which is used for model-based engineering and that generates code for robotics, Amesim by SIEMENS, which simulates the motor in context, and the S3D software, which analyses system performance. 

The project, which will finish in September 2022, also includes the roll-out of 11 demonstrations for use in five fields (transport, construction, logistics, inspection and surveillance, the latter for the agriculture industry in particular). All of them are currently being tested. “The project is reaching its growth and technology evaluation phase,” announces Reda Nouacer. 

 

The Art of Tolerating Flaws

We now know that improving the reliability of UAVs is a major area of concern. Nevertheless, like any other machine, a drone can suffer from failures. Among the elements on which it relies, is the flight control system, which ensures the stability and control of the drone. This is the purpose of the flaw tolerance diagnosis and control methods that are developed by the SIAM (Signal, Image, Automatism) team at the Computer Science, Bioinformatics and Complex Systems Laboratory (IBISC – Univ. Paris-Saclay, Univ. d’Évry). Lydie Nouvelière, Dalil Ichalal and their colleagues design algorithms that can estimate the presence of flaws and isolate them, calculate control laws that are tolerant of them and analyse their robustness. Ultimately, the models they develop help to define a strategy for maintaining the system.  

Firstly, the researchers analyse flight dynamics, which differ according to the UAV in question. This involves mathematically expressing the structure and behaviour of the UAV over time (speed, position, angles, orientation, etc.). It is often necessary to simplify equations. “We extract a simplified synthesis model to use it more easily while still trying to keep a certain realism,” explains Dalil Ichalal. To develop it, researchers use either predefined models – models with six degrees of freedom, forces and moments, dynamic coefficients – or AI-derived data. Other models – linear with variable parameters, for example – apply more specifically to situations where the drone’s geometry changes during the mission. This is the case for a drone whose mass and inertia vary with the delivery of the package, or for a drone equipped with a mobile arm. To go even further, the team is now looking to combine methods: “We will supplement the model with artificial intelligence to estimate the simplifications made initially,” emphasises Lydie Nouvelière. 

Then, they have to calculate a nominal flight control law. “We study its scope of application to verify that it is valid and capable of operating a trajectory,” explains Dalil Ichalal. They also study other issues that may arise during a drone’s flight, such as gales or obstacles (birds and pylons). “These external and unpredictable disruptions are unknown inputs for the system. They define a disrupted model and control law.” With a solid control law, the drone can estimate the disruption, and balance out its trajectory in real time. Disruptions are sometimes internal, such as communication loss due to a malfunctioning sensor. This is a ‘flaw’. “A flaw is abnormal functioning of the system. It is different from a disruption due to its permanent nature,” explains Dalil Ichalal. The most frequent flaws are linked to wear and tear on the actuators (jacks). Other are sporadic, such as loose contacts. “There are also system flaws, such as fractures, which changes the drone’s aerodynamism.” 

Once the flaw has been diagnosed, the system will determine the category into which it should be classed, in order to isolate it. Yet, in engineering, it is widespread practice to duplicate components or essential functions. Speed, for example, is measured using several, distinct types of sensors, and the sensitivity of measurements of a given flaw generates flaw characteristic patterns. “It is then a case of determining if the flaw is severe or not, and if we can reconfigure the control law to take the flaw into account. We can, for example, replace a sensor with an ‘observer’, which is an algorithm that can predict and estimate the missing measurements,” remarks Lydie Nouvelière. This is ‘flaw tolerant control’. Nevertheless, not all flaws are tolerable: the system must calculate to see if the feasibility conditions of the trajectory are still met, and to replan it if necessary. 

 

Management of the Fleet in Question

While it is critical to guarantee the flight security of the drone, operating it can be even more complicated when it means using a fleet of drones over an area that is to be explored, while searching for targets (debris, people, vehicles), whether they are static or not. The challenge, therefore, is to ensure they evolve jointly and without danger, so that they are capable of remotely notifying the presence of targets of interest and that they can differentiate them from areas without these targets. “Emergency services can reach areas faster or a patrol can avoid going into a combat area,” says Hélène Piet-Lahanier, from the Information Processing and Systems Department (DTIS) at ONERA Palaiseau. She and her colleague, Sylvain Bertrand, frequently collaborate with Cristina Stoica Maniu and Michel Kieffer’s teams, from the Laboratory of Signals and Systems (L2S – Université Paris-Saclay, CNRS, CentraleSupélec) on the issue of correct co-operation of a fleet. Several aspects are decisive: communication, management, and flight effectiveness. 

This co-operation can be seen in three ways. It can be centralised with one, unique operator, and therefore the entire fleet will share its gathered information with this operator. The operator will define a strategy that can be applied to all the drones, and when needed, reallocate missions using an optimisation algorithm. Reliable and constant communication is essential. Conversely, it can be a decentralised system, wherein each drone can conduct its mission without concerning itself about its neighbouring drones. A middle approach, that of a distributed system, sees that drones that are close to one another exchange collected information, depending on their possible means of communication. “The information is then sent to other drones outside the first emitting circle, and this creates local communication clusters,” describes Hélène Piet-Lahanier, who works on this type of method. It is undeniably a timesaver but is limited if the area is spread out and littered with obstacles. “The drones cannot exchange information between one another for a long while.” 

Implementing a mobility strategy that maximises high-quality data collection and that avoids collision is crucial. Certain elements, such as “the known presence of an obstacle, hypotheses about target movement and their absence in certain areas for specific reason,” can partially influence this. The type of UAV used determines another. “It is about defining a coherent trajectory. For example, a fixed-wing drone cannot hover, nor can it do a hairpin bed, unlike a quadcopter.” After this comes estimation algorithm definition. “It determines where one or more targets are present, and where none are present.” The crux of the problem is coordinating drone flights to ensure the whole area of interest is swept, and to get a clear point of view. “The area where the target may be located must be visualised from all angles to ensure detection, even in the presence of a cloaking device. If the target is mobile, it is more complex, as this visualisation must be simultaneous, and more restrictive in terms of fleet management.” Drones therefore will draw a specific map of the location of objects of interest, which is added to over time, as they explore. 

 

If Retreating is the Solution

Errors in assessment cannot be excluded. “Sometimes the drone does not have the right angle of view, or sensor, to remove any ambiguity, and mistakes a decoy for a target. It is by cross-referencing information collected by other drones that we can discriminate it.” In the same way, if a piece of collected information does not fit the general trends, it means that the drone has a problem. “Thanks to co-operation and exchange between drones, we can find the incoherent drone,” and thus it will be reallocated to a different mission. “We will allocate it to an area where it will not be a danger and where information is not decisive.” If it is no longer physically able to follow the movement, it will carry out a diagnosis of its possibilities of evolution, and according to the result, will land in emergency or open its parachute. 

Successfully retreating one or several drones from a fleet without any problems is the area of focus for Cristina Stoica Maniu, from the L2S, and her colleagues, who use predictive control. This advance control, which is most used in automation, is based on a mathematical model that can anticipate the behaviour of the fleet, which is comparable to a multi-agent system working in a defined area. Researchers have recently developed a new decentralised algorithm for the deployment and reconfiguration of a drone formation during a mission. The algorithm constantly marks out the zone in the form of a polygon tessellation, Voronoi cells, each of which has a localised drone in the Chebyshev centre, the centre of the largest enclosed circle in a polygon. This tessellation changes with the movement of the drones, and if a drone leaves the fleet, the drones will align themselves again on a new centre to avoid any collision. “The algorithm calculates the new distribution of Voronoi cells and Chebyshev centres to reorganise the configuration of the drone fleet.” Once the drone has retreated, the others will restart their initial task. 

Teams from IBISC, ONERA and L2S are currently optimising their algorithms and implementing them in drones, to conduct experiments within their respective drone flight arena. “After simulations, it is the only way to experience real conditions and expected performances,” concludes Lydie Nouvelière. 
 

 

 

Publications :

  • Radermacher A., et al. Designing Drone Systems with Papyrus for Robotics. In Proceedings of the 2021 Drone Systems Engineering and Rapid Simulation and Performance Evaluation: Methods and Tools Proceedings, (2021). 
  • Bouzgou K., et al. PD Sliding Mode Controller Based Decoupled Aerial Manipulation. ICINCO (2020). 
  • Abouselima E., et al., Robust Actuator Fault Diagnosis for LPV systems: Application to Quadrotor. American Control Conference, (2021). 
  • Ibenthal J., et al. Bounded-error target localization and tracking using a fleet of UAVs, Automatica, vol. 132, 109209, (2021). 
  • Chevet T., et al. Decentralized MPC for UAVs Formation Deployment and Reconfiguration with Multiple Outgoing Agents. J Intell Robot Syst 97, (2020). 

 

 

Can drones and the law take off together?

Although the law has suffered from a certain delay in the field of UAVs, it has been trying to make up for this over the past few years. 

Initially, the reference legal text in the field of international civil aviation, the Chicago Convention of 1944, only had one article dedicated to UAVs. Article 8 forbids UAV to fly over the territory of a foreign country, except with special authorisation. Generally, UAV flights across a border are always marginal. “At the time, we were preparing to end World War II, and the many possibilities of drones had yet to be considered,” says Vincent Correia, professor in public law at the Jean Monnet Faculty (Law, Economics and Management). “Additionally, these UAVs have a multitude of means of propulsion and weights and did not fall into any categories outlined in traditional aviation law (small aircraft, microlights, commercial aeroplanes, helicopters).” 

Following several incidents linked to drones nearing international airports, the International Civil Aviation Organization (ICAO) took up the issue. “The Annex 2 of the Chicago Convention was amended in 2012 to include the definition of UAV, as well as the Annex 13 on post-accident investigation.” At the same time, under Regulation 216/2008, the European Union Aviation Safety Agency (EASA) intervened for drones weighing more than 150 kg, which are far from being the majority. 

Faced with European shortcomings, national regulations took up the task, and France is leading the way. In 2012, the French Civil Aviation Authority (DGAC) adopted two decrees to provide guidance in the classification of civil airborne drones into categories of model aircraft, and remote-controlled and non-remotecontrolled aircraft. These categories combined maximum take-off mass, operating conditions, and pilot capability. In 2015, France adopted two new decrees to replace those from 2012. “The law was fine, but perhaps insufficiently detailed, and there was especially a problem in the application of the norm.” These new decrees added more restrictions and increased information and training obligations for pilots. They also distinguished between remotecontrolled aircraft and autonomous aircraft, and outlined rules for using autonomous and semi-autonomous systems. “We had, for the first time, defining elements as to what an autonomous flight was.” 

European Regulation 2018/1139, which was adopted in 2018, expanded EASA’s competences. It dropped the 150 kg limit and adopted a phased approach, comparable to the French rules. Adopted in 2019, the Implementing Regulation 2019/947 distinguished between three categories of flight: open (low risk), specific (medium risk) and certified (high risk), correlating flight scenarios and UAV mass. It specified rules for open category UAV regarding training obligations for pilots, including sometimes having to sit an exam. Delegated Regulation 2019/945 set out the design and manufacturing requirements for UAVs to be operated. 
European and national supervisory authorities have taken on the responsibility to uphold these rules. In France, it is the DGAC, the gendarmerie and the police who have this role. “National monitoring is applied according to criteria that are gradually being harmonised at European level. European and national law supplement one another, and in the event of incompatibility, the former takes precedence over the latter.” Regarding data and privacy protection, traditional instruments are applied (the French Civil Code, right to respect for private and family life, French Penal Code, etc.), which are usually sufficient. 

Automation of UAVs does nevertheless pose regulatory problems: in the event of a dispute, who is responsible if a teleoperator is absent? The manufacturer of the drone or the public authority that authorised the autonomous flight? “For the time being, the technology is not regarded as mature and the approach of the authorities remains very cautious.” Will drones one day be able to fly in unsegregated airspace, once reserved for civil aviation and authorised flights? “It will first be necessary to consider the already complex task of air traffic controllers and to allay the fears of the populations overflown by these aircraft.”