Pasar al contenido principal

SUPERVISION DE FLOTA DE DRONES Y OPTIMIZACION DE LOS PLANES DE VUELO DE OPERACIONES COMERCIALES

PID2020-116377RB-C21

•
Nombre agencia financiadora Agencia Estatal de Investigación
Acrónimo agencia financiadora AEI
Programa Programa Estatal de I+D+i Orientado a los Retos de la Sociedad
Convocatoria Proyectos de I+D+i (Generación de Conocimiento y Retos Investigación)
Año convocatoria 2020
Plan Estatal 2017-2020 Plan Estatal de Investigación Científica y Técnica y de Innovación
Centro beneficiario UNIVERSITAT POLITECNICA DE CATALUNYA

Resultados relacionados

Resultados totales (Incluyendo duplicados): 31
Encontrada(s) 1 página(s)

UPCommons. Portal del coneixement obert de la UPC
Publicación Artículo científico (JournalArticle)

Countering a drone in a 3D space: Analyzing deep reinforcement learning methods

  • Çetin, Ender|||0000-0002-9258-1919
  • Barrado Muxí, Cristina|||0000-0003-0100-724X
  • Pastor Llorens, Enric|||0000-0002-7587-8702
Unmanned aerial vehicles (UAV), also known as drones have been used for a variety of reasons and the commercial drone market growth is expected to reach remarkable levels in the near future. However, some drone users can mistakenly or intentionally fly into flight paths at major airports, flying too close to commercial aircraft or invading people’s privacy. In order to prevent these unwanted events, counter-drone technology is needed to eliminate threats from drones and hopefully they can be integrated into the skies safely. There are various counter-drone methods available in the industry. However, a counter-drone system supported by an artificial intelligence (AI) method can be an efficient way to fight against drones instead of human intervention. In this paper, a deep reinforcement learning (DRL) method has been proposed to counter a drone in a 3D space by using another drone. In a 2D space it is already shown that the deep reinforcement learning method is an effective way to counter a drone. However, countering a drone in a 3D space with another drone is a very challenging task considering the time required to train and avoid obstacles at the same time. A Deep Q-Network (DQN) algorithm with dueling network architecture and prioritized experience replay is presented to catch another drone in the environment provided by an Airsim simulator. The models have been trained and tested with different scenarios to analyze the learning progress of the drone. Experiences from previous training are also transferred before starting a new training by pre-processing the previous experiences and eliminating those considered as bad experiences. The results show that the best models are obtained with transfer learning and the drone learning progress has been increased dramatically. Additionally, an algorithm which combines imitation learning and reinforcement learning is implemented to catch the target drone. In this algorithm, called deep q-learning from demonstrations (DQfD), expert demonstrations data and self-generated data by the agent are sampled and the agent continues learning without overwriting the demonstration data. The main advantage of this algorithm is to accelerate the learning process even if there is a small amount of demonstration data., This work was funded partially by the AGAUR under grant 2020PANDE00141, the Ministry of Science and Innovation of Spain under grant PID2020-116377RB-C21 and the SESAR Joint Undertaking (JU) project CORUS-XUAM, under grant SESAR-VLD2 101017682. The JU receives support from the European Union’s Horizon 2020 research and innovation program and SESAR JU members other than the Union., Peer Reviewed
UPCommons. Portal del coneixement obert de la UPC
Publicación Contribución de congreso (ConferenceOutput)

Drone performance modelling for U-space strategic deconflicting services

  • Sánchez Segura, Albert|||0009-0008-3000-2871
  • Pastor Llorens, Enric|||0000-0002-7587-8702
  • Barrado Muxí, Cristina|||0000-0003-0100-724X
Integrating drones into U-space airspace demands accurate performance modelling to support strategic deconfliction and reliable 4D trajectory predictions. However, most commercially available drones lack standardised performance profiles, making current ATM-oriented models like BADA not directly usable for U-space services. This paper presents a kinematic, operationally-focused approach to drone performance modelling based on classical dynamic principles and real flight data from representative drone platforms. The methodology divides flight into key phases, such as turns, climbs, and transitions, to generate polynomial estimators that capture essential dynamic behaviour without requiring complex sub-system modelling. The proposed models are packaged for easy integration by Uspace service developers and validated against actual drone flight trajectories. Results show improved prediction accuracy, highlighting the value of tailored drone performance models in enabling safe and efficient U-space operations., This research was supported by grants PID2023-152053OB-C21 and PID2020-116377RB-C21, funded by MICIU/AEI /10.13039/501100011033 and by FEDER, UE., Peer Reviewed
UPCommons. Portal del coneixement obert de la UPC
Publicación Artículo científico (JournalArticle)

Enhancing drones for law enforcement and capacity monitoring at open large events

  • Royo Chic, Pablo|||0000-0002-8829-7528
  • Asenjo Carvajal, Àlex
  • Trujillo Gómez, Juan Manuel
  • Çetin, Ender|||0000-0002-9258-1919
  • Barrado Muxí, Cristina|||0000-0003-0100-724X
Police tasks related with law enforcement and citizen protection have gained a very useful asset in drones. Crowded demonstrations, large sporting events, or summer festivals are typical situations when aerial surveillance is necessary. The eyes in the sky are moving from the use of manned helicopters to drones due to costs, environmental impact, and discretion, resulting in local, regional, and national police forces possessing specific units equipped with drones. In this paper, we describe an artificial intelligence solution developed for the Castelldefels local police (Barcelona, Spain) to enhance the capabilities of drones used for the surveillance of large events. In particular, we propose a novel methodology for the efficient integration of deep learning algorithms in drone avionics. This integration improves the capabilities of the drone for tasks related with capacity control. These tasks have been very relevant during the pandemic and beyond. Controlling the number of persons in an open area is crucial when the expected crowd might exceed the capacity of the area and put humans in danger. The new methodology proposes an efficient and accurate execution of deep learning algorithms, which are usually highly demanding for computation resources. Results show that the state-of-the-art artificial intelligence models are too slow when utilised in the drone standard equipment. These models lose accuracy when images are taken at altitudes above 30 metres. With our new methodology, these two drawbacks can be overcome and results with good accuracy (96% correct segmentation and between 20% and 35% mean average proportional error) can be obtained in less than 20 s., This research was partially funded by the AGAUR research agency of Catalonia under grant number 2020PANDE00141 and by the Ministry of Science and Education of Spain under grant number PID2020-116377RB-C21., Peer Reviewed
UPCommons. Portal del coneixement obert de la UPC
Publicación Artículo científico (JournalArticle)

Analyzing deep reinforcement learning model decisions with Shapley additive explanations for counter drone operations

  • Çetin, Ender|||0000-0002-9258-1919
  • Barrado Muxí, Cristina|||0000-0003-0100-724X
  • Salamí San Juan, Esther|||0000-0002-4635-2963
  • Pastor Llorens, Enric|||0000-0002-7587-8702
As the use of drones continues to increase, their capabilities pose a threat to airspace safety when they are misused. Deploying AI models for intercepting these unwanted drones becomes crucial. However, these AI models, such as deep learning models, often operate as “black boxes”, making it hard to trust their decision-making system. This also affects end-users’ confidence in these AI systems. In this paper, the explainability of deep reinforcement learning is investigated and a deep reinforcement learning (DRL) method, double deep Q-network with dueling architecture and prioritized experience replay is applied to train the AI models. To make the AI model decisions more transparent and to understand the reasoning behind the AI decisions for counter-drone systems, Shapley Additive Explanations (SHAP) method is implemented. After training the DRL agent, experience replay is visualized, and the absolute SHAP values are calculated to explain the key factors that influence the deep reinforcement learning agent’s choices. The integration of DRL with explainable AI methods such as SHAP demonstrates significant potential for the advancement of robust and efficient counter-drone systems., Open Access funding provided thanks to the CRUE-CSIC agreement with Springer Nature. This work was funded by the Ministry of Science and Innovation of Spain under grant PID2020-116377RB- C21 for drone fleet management and optimization of commercial operations flight plans., Peer Reviewed
UPCommons. Portal del coneixement obert de la UPC
Publicación Artículo científico (JournalArticle)

Compact workspace decomposition based on a bottom-up approach

  • Skorobogatov, Georgy|||0000-0003-2536-1470
  • Calvo Belmonte, Toni
  • Barrado Muxí, Cristina|||0000-0003-0100-724X
  • Salamí San Juan, Esther|||0000-0002-4635-2963
In this paper, we present an algorithm that addresses the challenge of dividing a workspace among multiple UAVs. The workspace can be any convex or non-convex polygon and may contain holes of various shapes that represent no-fly zones. The UAVs can be heterogeneous, with different levels of autonomy, speed, and range. The goal of the workspace division is to obtain areas whose sizes are best matched to the capabilities of the UAVs while maximizing compactness. The algorithm decomposes the polygon representing the workspace into a triangular grid, followed by an iterative process of accumulating adjacent triangles while maximizing the compactness of the resulting regions. The performance of the algorithm and the quality of the partitions generated by the algorithm are compared to existing methods. Results show that this approach outperforms others in several metrics, achieving a 5% to 10% improvement in compactness, while maintaining reasonable performance, with a time overhead of up to approximately two seconds when splitting a polygon into ten parts., This work was supported in part by Ministerio de Economía, Industria y Competitividad de España under Award BES-2017-079798; and in part by the Ministerio de Ciencia e Innovación del Gobierno de España under Grant PID2020-116377RB-C21 and Grant PID2023-152053OB-C21., Peer Reviewed
UPCommons. Portal del coneixement obert de la UPC
Publicación Contribución de congreso (ConferenceOutput)

Explainability of deep reinforcement learning method with drones

  • Çetin, Ender|||0000-0002-9258-1919
  • Barrado Muxí, Cristina|||0000-0003-0100-724X
  • Pastor Llorens, Enric|||0000-0002-7587-8702
Recent advances in artificial intelligence (AI) technology demonstrated that AI algorithms are very powerful as AI models become more complex. As a result, the users and also the engineers who developed the AI algorithms have a hard time explaining how the AI model gives the specific result. This phenomenon is known as "black box" and affects end-users’ confidence in these AI systems. In this research, explainability of deep reinforcement learning is investigated for counter-drone systems. To counter a drone, a deep reinforcement learning method such as double deep Q-network with dueling architecture and prioritized experience replay is proposed. In counter-drone systems, catching the target as soon as possible is expected. Otherwise, the target can be gone in a short time. To understand how the agent performs more quickly and accurately, figures representing rewards, drone locations, crash positions, and the distribution of actions are analyzed and compared. For example, the positions of the drones in a successful episode during training can be analyzed by the actions the agent performed and the rewards in this episode. In addition, the actions agent took in episodes are compared with action frequencies during training and it is seen that at the end of the training, the agent selects the dominant actions throughout the training. However, at the beginning of the training, the distribution of actions is not correlated with the actions selected at the end. The results showed that the agent uses different flight paths by using different actions to catch the target drone in different episodes and different models. Finally, the generation of a saliency map is investigated to identify the critical regions in an input image which influences the predictions made by the DQN agent by evaluating the gradients of the model’s output with respect to both the image and scalar inputs., This work was funded by the Ministry of Science and Innovation of Spain under grant PID2020-116377RB-C21., Peer Reviewed
UPCommons. Portal del coneixement obert de la UPC
Publicación Contribución de congreso (ConferenceOutput)

Insights of the citizens perception during a 3-days outdoors flight demonstration

  • Barrado Muxí, Cristina|||0000-0003-0100-724X
  • Kuljanin, Jovana|||0000-0002-3380-262X
  • Salamí San Juan, Esther|||0000-0002-4635-2963
  • Pastor Llorens, Enric|||0000-0002-7587-8702
Citizens are getting used to seeing more and more drones flying in the sky, but this is still a sporadic situation and it is dif¿cult to experience it in urban environments. The reason is that in low-populated areas the safety regulations are more permissible than in populated areas. With the exception of drone shows, the operations that citizens may observe in urban areas are mainly the ones devoted to security or law enforcement. The CORUS-XUAM project ran six drone flight demonstrations of urban air mobility in seven different European countries. Flights were demonstrated in Sweden, Germany, Belgium, Italy, France and Spain. In the case of the Spanish demonstration, drones were flying over the beach of Castelldefels, an urban area in which citizens go for a promenade. This situation was ideal to capture opinions from the citizens not involved in the operations. In addition, the number of flights accounted for 72 delivery operations and lasted three days. As part of the exercise evaluation a number of surveys was collected. From a total of 77 surveys responded during these exercises, 37 corresponded to citizens’ opinions. During the flight demonstrations we also captured the sound pressure and recorded some drone noises. In this paper we will present the citizens’ responses related to noise and will try to find correlations with the captured sound pressure and noise recordings. The results of this analysis can be directly used in the MUSE project to further refine the noise parameters that are perceived as more annoying by the citizens., This paper is partially funded by the Ministry Science and Education of Spain, under fund grant PID2020-116377RB-C21, and is also part of the MUSE project that has received fund- ing from the SESAR 3 Joint Undertaking (SESAR 3 JU) under grant No 101114858. The JU receives support from the European Union’s Horizon Europe research and innovation prgramme and the SESAR 3 JU members other than the Union. The resuls and opinions re- flect only the authors’ view and the SESAR JU and the Spanish Ministry are not responsible of any use that may be made of the information it contains., Peer Reviewed
UPCommons. Portal del coneixement obert de la UPC
Publicación Contribución de congreso (ConferenceOutput)

Counter a drone and the performance analysis of deep reinforcement learning method and human pilot

  • Çetin, Ender|||0000-0002-9258-1919
  • Barrado Muxí, Cristina|||0000-0003-0100-724X
  • Pastor Llorens, Enric|||0000-0002-7587-8702
Artificial Intelligence (AI) has been used in different research areas in aerospace to create an intelligent system. Especially, an unmanned aerial vehicle (UAV), known as a drone, can be controlled by AI methods such as deep reinforcement learning (DRL) in different purposes. Drones with DRL become more intelligent and eventually they can be fully autonomous. In this paper, DRL method supported by real time object detection model is proposed to detect and catch a drone. Additionally, the results are analyzed by comparing the time to catch the target drone in seconds between DRL method, human pilot and an algorithm which directs the drone towards the target position without using any AI method or navigation and guidance method. The main idea is to catch a drone in an environment as fast as possible without crashing any obstacles inside the environment. In DRL method, the agent is a quadcopter drone and it is rewarded in each time step by the environment provided by Airsim flight simulator. Drone is trained to catch the target drone by using DRL model which is based on deep Q-Network algorithm. After training, the tests have been made by the agent drone with DRL model and human pilots to catch stationary and non-stationary target drone. The training and test results show that the agent drone learns to catch target drone which can be a stationary and a non-stationary. In addition. the agent avoids crashing any obstacles in the environment with a minimum success rate of 94%. Also, DRL model performance is compared with the human pilot performances and the agent with DRL model shows better time to catch the target drone. Human pilots struggle to control the drone by using remote controller when catching the target in simulation. However, the agent with DRL model is rarely missing the target when trying to catch the target., This work was funded partially by the AGAUR under grant 2020PANDE00141, the Ministry of Science and Innovation of Spain under grant PID2020-116377RB-C21 and the SESAR Joint Undertaking (JU) project CORUS-XUAM, under grant SESAR-VLD2 101017682. The JU receives support from the European Union’s Horizon 2020 research and innovation programme and the SESAR JU members other than the Union. Special thanks to Dr. Pablo Royo Chic for his assistance to our paper by piloting a drone to catch the target drone.
UPCommons. Portal del coneixement obert de la UPC
Publicación Artículo científico (JournalArticle)

Vulnerable road users and connected autonomous vehicles interaction: a survey

  • Reyes Muñoz, María Angélica|||0000-0002-8284-9934
  • Guerrero Ibañez, Juan
There is a group of users within the vehicular traffic ecosystem known as Vulnerable Road Users (VRUs). VRUs include pedestrians, cyclists, motorcyclists, among others. On the other hand, connected autonomous vehicles (CAVs) are a set of technologies that combines, on the one hand, communication technologies to stay always ubiquitous connected, and on the other hand, automated technologies to assist or replace the human driver during the driving process. Autonomous vehicles are being visualized as a viable alternative to solve road accidents providing a general safe environment for all the users on the road specifically to the most vulnerable. One of the problems facing autonomous vehicles is to generate mechanisms that facilitate their integration not only within the mobility environment, but also into the road society in a safe and efficient way. In this paper, we analyze and discuss how this integration can take place, reviewing the work that has been developed in recent years in each of the stages of the vehicle-human interaction, analyzing the challenges of vulnerable users and proposing solutions that contribute to solving these challenges., This work was partially funded by the Ministry of Economy, Industry, and Competitiveness of Spain under Grant: Supervision of drone fleet and optimization of commercial operations flight plans, PID2020-116377RB-C21., Peer Reviewed
UPCommons. Portal del coneixement obert de la UPC
Publicación Artículo científico (JournalArticle)

Identification of potential Urban Air Mobility (UAM) safety hazards from helicopter accident data: methodology and preliminary results

  • Pons Prats, Jordi|||0000-0002-4930-9135
  • Kuljanin, Jovana|||0000-0002-3380-262X
  • Netjasov, Fedja
  • Saenz de Tejada Millan, Jorge
Safety is paramount for aviation. The development of urban air mobility (UAM) is not excluded from consideration, however, while research generally focuses on technical and operational issues, few have been done regarding safety. It is clear that there are a lot of unknowns regarding either the design or operation of the UAM vehicles, so this research paper proposes to extract as much information as possible from helicopter accidents in urban environments. While no detailed technical information about UAM vehicles and operations is available, helicopters represent an interesting source of data. It is considered that the UAM vehicles with vertical take-off and landing capabilities present similar characteristics to helicopters. Building on a previous report by the authors [1], a set of accident reports from several safety agencies worldwide has been analysed. The reports have been shortlisted according to the relevance of the scenario; selecting those which happened in an urban environment or close to urban areas. Relevant hazards have been identified and assessed by experts in order to understand how they could translate to UAM operations. Identified hazards could serve for risk assessment of future UAM operations as well as for the development of risk mitigation measures or policies., The authors would like to express their gratitude to their colleagues from the University of Belgrade (UBFTTE), and the Technical University of Catalonia (UPC, ICARUS Research group), for their contribution to the expert assessment brainstorming session conducted as part of this research; special thanks to Dr Bojana Mirkovic, and Dr Emir Ganic UB-FTTE, and Dr Cristina Barrado, Dr Enric Pastor, and Prof. Xavier Prats, UPC. This research has been partially supported through the CIMNE Severo Ochoa Centre of Excellence (2019-2023) under the grant CEX2018-000797-S funded by MCIN/AEI/10.13039/501100011033. This work was partially funded by the Ministerio de Economia y Competitividad of Spain under contracts PID2020-116377RB-C21 and PID2023-152053OB-C21. The author Jordi Pons-Prats is a Serra Hunter Fellow., Peer Reviewed
UPCommons. Portal del coneixement obert de la UPC
Publicación Artículo científico (JournalArticle)

Image-based multi-agent reinforcement learning for demand–capacity balancing

  • Mas Pujol, Sergi|||0000-0003-3059-6610
  • Salamí San Juan, Esther|||0000-0002-4635-2963
  • Pastor Llorens, Enric|||0000-0002-7587-8702
Air traffic flow management (ATFM) is of crucial importance to the European Air Traffic Control System due to two factors: first, the impact of ATFM, including safety implications on ATC operations; second, the possible consequences of ATFM measures on both airports and airlines operations. Thus, the central flow management unit continually seeks to improve traffic flow management to reduce delays and congestion. In this work, we investigated the use of reinforcement learning (RL) methods to compute policies to solve demand–capacity imbalances (a.k.a. congestion) during the pre-tactical phase. To address cases where the expected demands exceed the airspace sector capacity, we considered agents representing flights who have to decide on ground delays jointly. To overcome scalability issues, we propose using raw pixel images as input, which can represent an arbitrary number of agents without changing the system’s architecture. This article compares deep Q-learning and deep deterministic policy gradient algorithms with different configurations. Experimental results, using real-world data for training and validation, confirm the effectiveness of our approach to resolving demand–capacity balancing problems, showing the robustness of the RL approach presented in this article., This work was funded by EUROCONTROL under Ph.D. Research contract no. 18-220569- C2 and by the Ministry of Economy, Industry, and Competitiveness of Spain under grant number PID2020-116377RB-C21., Peer Reviewed
UPCommons. Portal del coneixement obert de la UPC
Publicación Artículo científico (JournalArticle)

Implementing mitigations for improving societal acceptance of urban air mobility

  • Çetin, Ender|||0000-0002-9258-1919
  • Cano Gento, Alicia
  • Deransy, Robin
  • Tres Martínez, Sergi
  • Barrado Muxí, Cristina|||0000-0003-0100-724X
The continuous development of technical innovations provides the opportunity to create new economic markets and a wealth of new services. However, these innovations sometimes raise concerns, notably in terms of societal, safety, and environmental impacts. This is the case for services related to the operation of unmanned aerial vehicles (UAV), which are emerging rapidly. Unmanned aerial vehicles, also called drones, date back to the first third of the twentieth century in aviation industry, when they were mostly used for military purposes. Nowadays, drones of various types and sizes are used for many purposes, such as precision agriculture, search and rescue missions, aerial photography, shipping and delivery, etc. Starting to operate in areas with low population density, drones are now looking for business in urban and suburban areas, in what is called urban air mobility (UAM). However, this rapid growth of the drone industry creates psychological fear of the unknown in some parts of society. Reducing this fear will play an important role in public acceptance of drone operations in urban areas. This paper presents the main concerns of society with regard to drone operations, as already captured in some public surveys, and proposes a list of mitigation measures to reduce these concerns. The proposed list is then analyzed, and its applicability to individual, urban, very large demonstration flights is explained, using the feedback from the CORUS-XUAM project. CORUS-XUAM will organize a set of very large drone flight demonstrations across seven European countries to investigate how to safely integrate drone operations into airspace with the support of the U-space., This project has received funding from the Ministry of Science and Innovation of Spain under grant PID2020-116377RB-C21 and from the SESAR Joint Undertaking (JU) under grant agreement No 101017682. The JU receives support from the European Union’s Horizon 2020 research and innovation programme and the SESAR JU members other than the Union., Peer Reviewed
UPCommons. Portal del coneixement obert de la UPC
Publicación Artículo científico (JournalArticle)

Assistive self-driving car networks to provide safe road ecosystems for disabled road users

  • Guerrero Ibañez, Juan Antonio
  • Contreras Castillo,, Juan José
  • Amezcua Valdovinos, Ismael
  • Reyes Muñoz, María Angélica|||0000-0002-8284-9934
Disabled pedestrians are among the most vulnerable groups in road traffic. Using technology to assist this vulnerable group could be instrumental in reducing the mobility challenges they face daily. On the one hand, the automotive industry is focusing its efforts on car automation. On the other hand, in recent years, assistive technology has been promoted as a tool for consolidating the functional independence of people with disabilities. However, the success of these technologies depends on how well they help self-driving cars interact with disabled pedestrians. This paper proposes an architecture to facilitate interaction between disabled pedestrians and self-driving cars based on deep learning and 802.11p wireless technology. Through the application of assistive technology, we can locate the pedestrian with a disability within the road traffic ecosystem, and we define a set of functionalities for the identification of hand gestures of people with disabilities. These functions enable pedestrians with disabilities to express their intentions, improving their confidence and safety level in tasks within the road ecosystem, such as crossing the street., This research was partially funded by the project “Drone fleet monitoring and optimization of commercial operations flight plans” PID2020-116377RB-C21. Research Challenges: R&D&I Projects, Spanish Research Agency., Peer Reviewed, Objectius de Desenvolupament Sostenible::10 - Reducció de les Desigualtats
UPCommons. Portal del coneixement obert de la UPC
Publicación Artículo científico (JournalArticle)

ATC human factors involved in RPAS contingency management in non-segregated airspace

  • Reyes Muñoz, María Angélica|||0000-0002-8284-9934
  • Barrado Muxí, Cristina|||0000-0003-0100-724X
  • Pastor Llorens, Enric|||0000-0002-7587-8702
  • Royo Chic, Pablo|||0000-0002-8829-7528
The overall approach towards Remotely Piloted Aerial System integration into a non-segregated airspace is that the unmanned vehicles should be able to fit into the current air traffic management system, thus meeting all the technical and regulatory requirements to be treated similar to any other airspace user. Such a requirement implies that unmanned aircraft operations should behave as close as possible to manned aviation or at least generate the most negligible possible negative impact on the system. From the air traffic management point of view, this implies that air traffic controllers should be capable of effectively handling different types of RPAS operating in a nominal state but also when suffering a potential contingency. This paper aims to analyse how air traffic controllers involved in managing unmanned aircraft integration into non-segregated airspace are impacted when an unmanned vehicle suffers a contingency. Participants: Six air traffic controllers were the test subjects, complemented by one RPAS pilot and several pseudo-pilots controlling the simulated manned traffic. The project collected real-time simulation data to develop specific indicators to determine how the controllers’ workload increases while managing complex traffic scenarios, including a single RPAS. Study Method: We conducted exhaustive traffic flight simulations, recreating complex airspace scenarios, including various RPAS types and mission-oriented trajectories. The involved RPAS were subjected to two of the most relevant contingencies: loss of the command-and-control link and engine failure. The experiments were evaluated in different operational scenarios, including using autonomous communication technologies to help air traffic controllers track the RPAS operation. Findings: The results indicate that the air traffic controller’s perception and workload are not affected beyond reason by the introduction of an unmanned aircraft as a new element into the non-segregated airspace, even when that aircraft suffers a contingency. The flight-intent technology increases situational awareness, leading to more efficient and safe airspace management. Additional simulations may need to be performed to evaluate the impact on airspace capacity, safety, and workload when various unmanned vehicles are simultaneously inserted., This research was funded by Ministry of Science and Innovation of Spain under Grant Number PID2020-116377RB-C21 and by Eurocontrol under Grant Number EU-08-120916-C, Peer Reviewed
UPCommons. Portal del coneixement obert de la UPC
Publicación Artículo científico (JournalArticle)

ACCORD: A formal model for the digitalization and automation of drone coordination processes

  • Pastor Llorens, Enric|||0000-0002-7587-8702
  • Macias López, Miquel
  • Martin Salas, Yeray Manel|||0009-0008-4936-9964
  • Sánchez Segura, Albert|||0009-0008-3000-2871
  • Barrado Muxí, Cristina|||0000-0003-0100-724X
This paper introduces ACCORD, a support platform designed to digitalize and automate the coordination processes required by the current drone regulatory framework. Drone operators must complete several coordination actions with both aeronautical and non-aeronautical entities. Traditional aeronautical coordination actions relate to the need to access protected airspace volumes around airports. Additional coordination should be established with smaller aeronautical infrastructures, like small aerodromes and heliports, which are not surrounded by any type of pre-defined airspace. Therefore, drone-specific protection volumes have been created. ACCORD enables a single entry point for all the necessary coordination processes for drone operators and infrastructure managers. The objective is to minimize the number of required actions, guarantee full traceability of the process, maximize access to the relevant information, automate the processes as much as possible, and maintain a high level of flexibility to support all coordination processes. After coordination is established, it moves from the strategic/planning phase to the actual execution phase. ACCORD also enables a communication mechanism between the drone operators and the aeronautical infrastructures to extend the coordination to the actual mission execution. ACCORD is currently being tested by some of the most relevant actors in the Catalan drone ecosystem. The current version of the system provides support for all types of aeronautical infrastructures (heliports, aerodromes, and airports) and management duality for situations in which the infrastructure manager and the aeronautical service provider coexist., This research was supported by grants PID2023-152053OB-C21 and PID2020-116377RB-C21, funded by MICIU/AEI /10.13039/501100011033 and by FEDER, UE.
UPCommons. Portal del coneixement obert de la UPC
Publicación Artículo científico (JournalArticle)

Monitoring tomato leaf disease through convolutional neural networks

  • Guerrero Ibañez, Juan Antonio
  • Reyes Muñoz, María Angélica|||0000-0002-8284-9934
Agriculture plays an essential role in Mexico’s economy. The agricultural sector has a 2.5% share of Mexico’s gross domestic product. Specifically, tomatoes have become the country’s most exported agricultural product. That is why there is an increasing need to improve crop yields. One of the elements that can considerably affect crop productivity is diseases caused by agents such as bacteria, fungi, and viruses. However, the process of disease identification can be costly and, in many cases, time-consuming. Deep learning techniques have begun to be applied in the process of plant disease identification with promising results. In this paper, we propose a model based on convolutional neural networks to identify and classify tomato leaf diseases using a public dataset and complementing it with other photographs taken in the fields of the country. To avoid overfitting, generative adversarial networks were used to generate samples with the same characteristics as the training data. The results show that the proposed model achieves a high performance in the process of detection and classification of diseases in tomato leaves: the accuracy achieved is greater than 99% in both the training dataset and the test dataset., This work was partially funded by the State Research Agency of Spain under grant number PID2020-116377RB-C21., Peer Reviewed
Recercat. Dipósit de la Recerca de Catalunya
Publicación Artículo científico (JournalArticle)

Implementing mitigations for improving societal acceptance of urban air mobility

  • Çetin, Ender
  • Cano Gento, Alicia
  • Deransy, Robin
  • Tres Martínez, Sergi
  • Barrado Muxí, Cristina
The continuous development of technical innovations provides the opportunity to create new economic markets and a wealth of new services. However, these innovations sometimes raise concerns, notably in terms of societal, safety, and environmental impacts. This is the case for services related to the operation of unmanned aerial vehicles (UAV), which are emerging rapidly. Unmanned aerial vehicles, also called drones, date back to the first third of the twentieth century in aviation industry, when they were mostly used for military purposes. Nowadays, drones of various types and sizes are used for many purposes, such as precision agriculture, search and rescue missions, aerial photography, shipping and delivery, etc. Starting to operate in areas with low population density, drones are now looking for business in urban and suburban areas, in what is called urban air mobility (UAM). However, this rapid growth of the drone industry creates psychological fear of the unknown in some parts of society. Reducing this fear will play an important role in public acceptance of drone operations in urban areas. This paper presents the main concerns of society with regard to drone operations, as already captured in some public surveys, and proposes a list of mitigation measures to reduce these concerns. The proposed list is then analyzed, and its applicability to individual, urban, very large demonstration flights is explained, using the feedback from the CORUS-XUAM project. CORUS-XUAM will organize a set of very large drone flight demonstrations across seven European countries to investigate how to safely integrate drone operations into airspace with the support of the U-space., This project has received funding from the Ministry of Science and Innovation of Spain under grant PID2020-116377RB-C21 and from the SESAR Joint Undertaking (JU) under grant agreement No 101017682. The JU receives support from the European Union’s Horizon 2020 research and innovation programme and the SESAR JU members other than the Union., Peer Reviewed, Postprint (published version)
Recercat. Dipósit de la Recerca de Catalunya
Publicación Artículo científico (JournalArticle)

Vulnerable road users and connected autonomous vehicles interaction: a survey

  • Reyes Muñoz, María Angélica
  • Guerrero Ibañez, Juan
There is a group of users within the vehicular traffic ecosystem known as Vulnerable Road Users (VRUs). VRUs include pedestrians, cyclists, motorcyclists, among others. On the other hand, connected autonomous vehicles (CAVs) are a set of technologies that combines, on the one hand, communication technologies to stay always ubiquitous connected, and on the other hand, automated technologies to assist or replace the human driver during the driving process. Autonomous vehicles are being visualized as a viable alternative to solve road accidents providing a general safe environment for all the users on the road specifically to the most vulnerable. One of the problems facing autonomous vehicles is to generate mechanisms that facilitate their integration not only within the mobility environment, but also into the road society in a safe and efficient way. In this paper, we analyze and discuss how this integration can take place, reviewing the work that has been developed in recent years in each of the stages of the vehicle-human interaction, analyzing the challenges of vulnerable users and proposing solutions that contribute to solving these challenges., This work was partially funded by the Ministry of Economy, Industry, and Competitiveness of Spain under Grant: Supervision of drone fleet and optimization of commercial operations flight plans, PID2020-116377RB-C21., Peer Reviewed, Postprint (published version)
Recercat. Dipósit de la Recerca de Catalunya
Publicación Artículo científico (JournalArticle)

Enhancing drones for law enforcement and capacity monitoring at open large events

  • Royo Chic, Pablo
  • Asenjo Carvajal, Àlex
  • Trujillo Gómez, Juan Manuel
  • Çetin, Ender
  • Barrado Muxí, Cristina
Police tasks related with law enforcement and citizen protection have gained a very useful asset in drones. Crowded demonstrations, large sporting events, or summer festivals are typical situations when aerial surveillance is necessary. The eyes in the sky are moving from the use of manned helicopters to drones due to costs, environmental impact, and discretion, resulting in local, regional, and national police forces possessing specific units equipped with drones. In this paper, we describe an artificial intelligence solution developed for the Castelldefels local police (Barcelona, Spain) to enhance the capabilities of drones used for the surveillance of large events. In particular, we propose a novel methodology for the efficient integration of deep learning algorithms in drone avionics. This integration improves the capabilities of the drone for tasks related with capacity control. These tasks have been very relevant during the pandemic and beyond. Controlling the number of persons in an open area is crucial when the expected crowd might exceed the capacity of the area and put humans in danger. The new methodology proposes an efficient and accurate execution of deep learning algorithms, which are usually highly demanding for computation resources. Results show that the state-of-the-art artificial intelligence models are too slow when utilised in the drone standard equipment. These models lose accuracy when images are taken at altitudes above 30 metres. With our new methodology, these two drawbacks can be overcome and results with good accuracy (96% correct segmentation and between 20% and 35% mean average proportional error) can be obtained in less than 20 s., This research was partially funded by the AGAUR research agency of Catalonia under grant number 2020PANDE00141 and by the Ministry of Science and Education of Spain under grant number PID2020-116377RB-C21., Peer Reviewed, Postprint (published version)
Recercat. Dipósit de la Recerca de Catalunya
Publicación Artículo científico (JournalArticle)

Countering a drone in a 3D space: Analyzing deep reinforcement learning methods

  • Çetin, Ender
  • Barrado Muxí, Cristina
  • Pastor Llorens, Enric
Unmanned aerial vehicles (UAV), also known as drones have been used for a variety of reasons and the commercial drone market growth is expected to reach remarkable levels in the near future. However, some drone users can mistakenly or intentionally fly into flight paths at major airports, flying too close to commercial aircraft or invading people’s privacy. In order to prevent these unwanted events, counter-drone technology is needed to eliminate threats from drones and hopefully they can be integrated into the skies safely. There are various counter-drone methods available in the industry. However, a counter-drone system supported by an artificial intelligence (AI) method can be an efficient way to fight against drones instead of human intervention. In this paper, a deep reinforcement learning (DRL) method has been proposed to counter a drone in a 3D space by using another drone. In a 2D space it is already shown that the deep reinforcement learning method is an effective way to counter a drone. However, countering a drone in a 3D space with another drone is a very challenging task considering the time required to train and avoid obstacles at the same time. A Deep Q-Network (DQN) algorithm with dueling network architecture and prioritized experience replay is presented to catch another drone in the environment provided by an Airsim simulator. The models have been trained and tested with different scenarios to analyze the learning progress of the drone. Experiences from previous training are also transferred before starting a new training by pre-processing the previous experiences and eliminating those considered as bad experiences. The results show that the best models are obtained with transfer learning and the drone learning progress has been increased dramatically. Additionally, an algorithm which combines imitation learning and reinforcement learning is implemented to catch the target drone. In this algorithm, called deep q-learning from demonstrations (DQfD), expert demonstrations data and self-generated data by the agent are sampled and the agent continues learning without overwriting the demonstration data. The main advantage of this algorithm is to accelerate the learning process even if there is a small amount of demonstration data., This work was funded partially by the AGAUR under grant 2020PANDE00141, the Ministry of Science and Innovation of Spain under grant PID2020-116377RB-C21 and the SESAR Joint Undertaking (JU) project CORUS-XUAM, under grant SESAR-VLD2 101017682. The JU receives support from the European Union’s Horizon 2020 research and innovation program and SESAR JU members other than the Union., Peer Reviewed, Postprint (published version)
Recercat. Dipósit de la Recerca de Catalunya
Publicación Contribución de congreso (ConferenceOutput)

Counter a drone and the performance analysis of deep reinforcement learning method and human pilot

  • Çetin, Ender
  • Barrado Muxí, Cristina
  • Pastor Llorens, Enric
Artificial Intelligence (AI) has been used in different research areas in aerospace to create an intelligent system. Especially, an unmanned aerial vehicle (UAV), known as a drone, can be controlled by AI methods such as deep reinforcement learning (DRL) in different purposes. Drones with DRL become more intelligent and eventually they can be fully autonomous. In this paper, DRL method supported by real time object detection model is proposed to detect and catch a drone. Additionally, the results are analyzed by comparing the time to catch the target drone in seconds between DRL method, human pilot and an algorithm which directs the drone towards the target position without using any AI method or navigation and guidance method. The main idea is to catch a drone in an environment as fast as possible without crashing any obstacles inside the environment. In DRL method, the agent is a quadcopter drone and it is rewarded in each time step by the environment provided by Airsim flight simulator. Drone is trained to catch the target drone by using DRL model which is based on deep Q-Network algorithm. After training, the tests have been made by the agent drone with DRL model and human pilots to catch stationary and non-stationary target drone. The training and test results show that the agent drone learns to catch target drone which can be a stationary and a non-stationary. In addition. the agent avoids crashing any obstacles in the environment with a minimum success rate of 94%. Also, DRL model performance is compared with the human pilot performances and the agent with DRL model shows better time to catch the target drone. Human pilots struggle to control the drone by using remote controller when catching the target in simulation. However, the agent with DRL model is rarely missing the target when trying to catch the target., This work was funded partially by the AGAUR under grant 2020PANDE00141, the Ministry of Science and Innovation of Spain under grant PID2020-116377RB-C21 and the SESAR Joint Undertaking (JU) project CORUS-XUAM, under grant SESAR-VLD2 101017682. The JU receives support from the European Union’s Horizon 2020 research and innovation programme and the SESAR JU members other than the Union. Special thanks to Dr. Pablo Royo Chic for his assistance to our paper by piloting a drone to catch the target drone., Postprint (published version)
Recercat. Dipósit de la Recerca de Catalunya
Publicación Artículo científico (JournalArticle)

Image-based multi-agent reinforcement learning for demand–capacity balancing

  • Mas Pujol, Sergi
  • Salamí San Juan, Esther
  • Pastor Llorens, Enric
Air traffic flow management (ATFM) is of crucial importance to the European Air Traffic Control System due to two factors: first, the impact of ATFM, including safety implications on ATC operations; second, the possible consequences of ATFM measures on both airports and airlines operations. Thus, the central flow management unit continually seeks to improve traffic flow management to reduce delays and congestion. In this work, we investigated the use of reinforcement learning (RL) methods to compute policies to solve demand–capacity imbalances (a.k.a. congestion) during the pre-tactical phase. To address cases where the expected demands exceed the airspace sector capacity, we considered agents representing flights who have to decide on ground delays jointly. To overcome scalability issues, we propose using raw pixel images as input, which can represent an arbitrary number of agents without changing the system’s architecture. This article compares deep Q-learning and deep deterministic policy gradient algorithms with different configurations. Experimental results, using real-world data for training and validation, confirm the effectiveness of our approach to resolving demand–capacity balancing problems, showing the robustness of the RL approach presented in this article., This work was funded by EUROCONTROL under Ph.D. Research contract no. 18-220569- C2 and by the Ministry of Economy, Industry, and Competitiveness of Spain under grant number PID2020-116377RB-C21., Peer Reviewed, Postprint (published version)
Recercat. Dipósit de la Recerca de Catalunya
Publicación Artículo científico (JournalArticle)

Assistive self-driving car networks to provide safe road ecosystems for disabled road users

  • Guerrero Ibañez, Juan Antonio
  • Contreras Castillo,, Juan José
  • Amezcua Valdovinos, Ismael
  • Reyes Muñoz, María Angélica
Disabled pedestrians are among the most vulnerable groups in road traffic. Using technology to assist this vulnerable group could be instrumental in reducing the mobility challenges they face daily. On the one hand, the automotive industry is focusing its efforts on car automation. On the other hand, in recent years, assistive technology has been promoted as a tool for consolidating the functional independence of people with disabilities. However, the success of these technologies depends on how well they help self-driving cars interact with disabled pedestrians. This paper proposes an architecture to facilitate interaction between disabled pedestrians and self-driving cars based on deep learning and 802.11p wireless technology. Through the application of assistive technology, we can locate the pedestrian with a disability within the road traffic ecosystem, and we define a set of functionalities for the identification of hand gestures of people with disabilities. These functions enable pedestrians with disabilities to express their intentions, improving their confidence and safety level in tasks within the road ecosystem, such as crossing the street., This research was partially funded by the project “Drone fleet monitoring and optimization of commercial operations flight plans” PID2020-116377RB-C21. Research Challenges: R&D&I Projects, Spanish Research Agency., Peer Reviewed, Objectius de Desenvolupament Sostenible::10 - Reducció de les Desigualtats, Postprint (published version)
Recercat. Dipósit de la Recerca de Catalunya
Publicación Artículo científico (JournalArticle)

Identification of potential Urban Air Mobility (UAM) safety hazards from helicopter accident data: methodology and preliminary results

  • Pons Prats, Jordi
  • Kuljanin, Jovana
  • Netjasov, Fedja
  • Saenz de Tejada Millan, Jorge
Safety is paramount for aviation. The development of urban air mobility (UAM) is not excluded from consideration, however, while research generally focuses on technical and operational issues, few have been done regarding safety. It is clear that there are a lot of unknowns regarding either the design or operation of the UAM vehicles, so this research paper proposes to extract as much information as possible from helicopter accidents in urban environments. While no detailed technical information about UAM vehicles and operations is available, helicopters represent an interesting source of data. It is considered that the UAM vehicles with vertical take-off and landing capabilities present similar characteristics to helicopters. Building on a previous report by the authors [1], a set of accident reports from several safety agencies worldwide has been analysed. The reports have been shortlisted according to the relevance of the scenario; selecting those which happened in an urban environment or close to urban areas. Relevant hazards have been identified and assessed by experts in order to understand how they could translate to UAM operations. Identified hazards could serve for risk assessment of future UAM operations as well as for the development of risk mitigation measures or policies., The authors would like to express their gratitude to their colleagues from the University of Belgrade (UBFTTE), and the Technical University of Catalonia (UPC, ICARUS Research group), for their contribution to the expert assessment brainstorming session conducted as part of this research; special thanks to Dr Bojana Mirkovic, and Dr Emir Ganic UB-FTTE, and Dr Cristina Barrado, Dr Enric Pastor, and Prof. Xavier Prats, UPC. This research has been partially supported through the CIMNE Severo Ochoa Centre of Excellence (2019-2023) under the grant CEX2018-000797-S funded by MCIN/AEI/10.13039/501100011033. This work was partially funded by the Ministerio de Economia y Competitividad of Spain under contracts PID2020-116377RB-C21 and PID2023-152053OB-C21. The author Jordi Pons-Prats is a Serra Hunter Fellow., Peer Reviewed, Postprint (published version)
Recercat. Dipósit de la Recerca de Catalunya
Publicación Contribución de congreso (ConferenceOutput)

Insights of the citizens perception during a 3-days outdoors flight demonstration

  • Barrado Muxí, Cristina
  • Kuljanin, Jovana
  • Salamí San Juan, Esther
  • Pastor Llorens, Enric
Citizens are getting used to seeing more and more drones flying in the sky, but this is still a sporadic situation and it is dif¿cult to experience it in urban environments. The reason is that in low-populated areas the safety regulations are more permissible than in populated areas. With the exception of drone shows, the operations that citizens may observe in urban areas are mainly the ones devoted to security or law enforcement. The CORUS-XUAM project ran six drone flight demonstrations of urban air mobility in seven different European countries. Flights were demonstrated in Sweden, Germany, Belgium, Italy, France and Spain. In the case of the Spanish demonstration, drones were flying over the beach of Castelldefels, an urban area in which citizens go for a promenade. This situation was ideal to capture opinions from the citizens not involved in the operations. In addition, the number of flights accounted for 72 delivery operations and lasted three days. As part of the exercise evaluation a number of surveys was collected. From a total of 77 surveys responded during these exercises, 37 corresponded to citizens’ opinions. During the flight demonstrations we also captured the sound pressure and recorded some drone noises. In this paper we will present the citizens’ responses related to noise and will try to find correlations with the captured sound pressure and noise recordings. The results of this analysis can be directly used in the MUSE project to further refine the noise parameters that are perceived as more annoying by the citizens., This paper is partially funded by the Ministry Science and Education of Spain, under fund grant PID2020-116377RB-C21, and is also part of the MUSE project that has received fund- ing from the SESAR 3 Joint Undertaking (SESAR 3 JU) under grant No 101114858. The JU receives support from the European Union’s Horizon Europe research and innovation prgramme and the SESAR 3 JU members other than the Union. The resuls and opinions re- flect only the authors’ view and the SESAR JU and the Spanish Ministry are not responsible of any use that may be made of the information it contains., Peer Reviewed, Postprint (published version)
Recercat. Dipósit de la Recerca de Catalunya
Publicación Artículo científico (JournalArticle)

ATC human factors involved in RPAS contingency management in non-segregated airspace

  • Reyes Muñoz, María Angélica
  • Barrado Muxí, Cristina
  • Pastor Llorens, Enric
  • Royo Chic, Pablo
The overall approach towards Remotely Piloted Aerial System integration into a non-segregated airspace is that the unmanned vehicles should be able to fit into the current air traffic management system, thus meeting all the technical and regulatory requirements to be treated similar to any other airspace user. Such a requirement implies that unmanned aircraft operations should behave as close as possible to manned aviation or at least generate the most negligible possible negative impact on the system. From the air traffic management point of view, this implies that air traffic controllers should be capable of effectively handling different types of RPAS operating in a nominal state but also when suffering a potential contingency. This paper aims to analyse how air traffic controllers involved in managing unmanned aircraft integration into non-segregated airspace are impacted when an unmanned vehicle suffers a contingency. Participants: Six air traffic controllers were the test subjects, complemented by one RPAS pilot and several pseudo-pilots controlling the simulated manned traffic. The project collected real-time simulation data to develop specific indicators to determine how the controllers’ workload increases while managing complex traffic scenarios, including a single RPAS. Study Method: We conducted exhaustive traffic flight simulations, recreating complex airspace scenarios, including various RPAS types and mission-oriented trajectories. The involved RPAS were subjected to two of the most relevant contingencies: loss of the command-and-control link and engine failure. The experiments were evaluated in different operational scenarios, including using autonomous communication technologies to help air traffic controllers track the RPAS operation. Findings: The results indicate that the air traffic controller’s perception and workload are not affected beyond reason by the introduction of an unmanned aircraft as a new element into the non-segregated airspace, even when that aircraft suffers a contingency. The flight-intent technology increases situational awareness, leading to more efficient and safe airspace management. Additional simulations may need to be performed to evaluate the impact on airspace capacity, safety, and workload when various unmanned vehicles are simultaneously inserted., This research was funded by Ministry of Science and Innovation of Spain under Grant Number PID2020-116377RB-C21 and by Eurocontrol under Grant Number EU-08-120916-C, Peer Reviewed, Postprint (published version)
Recercat. Dipósit de la Recerca de Catalunya
Publicación Artículo científico (JournalArticle)

ACCORD: A formal model for the digitalization and automation of drone coordination processes

  • Pastor Llorens, Enric
  • Macias López, Miquel
  • Martin Salas, Yeray Manel
  • Sánchez Segura, Albert
  • Barrado Muxí, Cristina
This paper introduces ACCORD, a support platform designed to digitalize and automate the coordination processes required by the current drone regulatory framework. Drone operators must complete several coordination actions with both aeronautical and non-aeronautical entities. Traditional aeronautical coordination actions relate to the need to access protected airspace volumes around airports. Additional coordination should be established with smaller aeronautical infrastructures, like small aerodromes and heliports, which are not surrounded by any type of pre-defined airspace. Therefore, drone-specific protection volumes have been created. ACCORD enables a single entry point for all the necessary coordination processes for drone operators and infrastructure managers. The objective is to minimize the number of required actions, guarantee full traceability of the process, maximize access to the relevant information, automate the processes as much as possible, and maintain a high level of flexibility to support all coordination processes. After coordination is established, it moves from the strategic/planning phase to the actual execution phase. ACCORD also enables a communication mechanism between the drone operators and the aeronautical infrastructures to extend the coordination to the actual mission execution. ACCORD is currently being tested by some of the most relevant actors in the Catalan drone ecosystem. The current version of the system provides support for all types of aeronautical infrastructures (heliports, aerodromes, and airports) and management duality for situations in which the infrastructure manager and the aeronautical service provider coexist., This research was supported by grants PID2023-152053OB-C21 and PID2020-116377RB-C21, funded by MICIU/AEI /10.13039/501100011033 and by FEDER, UE., Postprint (published version)
Recercat. Dipósit de la Recerca de Catalunya
Publicación Contribución de congreso (ConferenceOutput)

Explainability of deep reinforcement learning method with drones

  • Çetin, Ender
  • Barrado Muxí, Cristina
  • Pastor Llorens, Enric
Recent advances in artificial intelligence (AI) technology demonstrated that AI algorithms are very powerful as AI models become more complex. As a result, the users and also the engineers who developed the AI algorithms have a hard time explaining how the AI model gives the specific result. This phenomenon is known as "black box" and affects end-users’ confidence in these AI systems. In this research, explainability of deep reinforcement learning is investigated for counter-drone systems. To counter a drone, a deep reinforcement learning method such as double deep Q-network with dueling architecture and prioritized experience replay is proposed. In counter-drone systems, catching the target as soon as possible is expected. Otherwise, the target can be gone in a short time. To understand how the agent performs more quickly and accurately, figures representing rewards, drone locations, crash positions, and the distribution of actions are analyzed and compared. For example, the positions of the drones in a successful episode during training can be analyzed by the actions the agent performed and the rewards in this episode. In addition, the actions agent took in episodes are compared with action frequencies during training and it is seen that at the end of the training, the agent selects the dominant actions throughout the training. However, at the beginning of the training, the distribution of actions is not correlated with the actions selected at the end. The results showed that the agent uses different flight paths by using different actions to catch the target drone in different episodes and different models. Finally, the generation of a saliency map is investigated to identify the critical regions in an input image which influences the predictions made by the DQN agent by evaluating the gradients of the model’s output with respect to both the image and scalar inputs., This work was funded by the Ministry of Science and Innovation of Spain under grant PID2020-116377RB-C21., Peer Reviewed, Postprint (author's final draft)
Recercat. Dipósit de la Recerca de Catalunya
Publicación Artículo científico (JournalArticle)

Analyzing deep reinforcement learning model decisions with Shapley additive explanations for counter drone operations

  • Çetin, Ender
  • Barrado Muxí, Cristina
  • Salamí San Juan, Esther
  • Pastor Llorens, Enric
As the use of drones continues to increase, their capabilities pose a threat to airspace safety when they are misused. Deploying AI models for intercepting these unwanted drones becomes crucial. However, these AI models, such as deep learning models, often operate as “black boxes”, making it hard to trust their decision-making system. This also affects end-users’ confidence in these AI systems. In this paper, the explainability of deep reinforcement learning is investigated and a deep reinforcement learning (DRL) method, double deep Q-network with dueling architecture and prioritized experience replay is applied to train the AI models. To make the AI model decisions more transparent and to understand the reasoning behind the AI decisions for counter-drone systems, Shapley Additive Explanations (SHAP) method is implemented. After training the DRL agent, experience replay is visualized, and the absolute SHAP values are calculated to explain the key factors that influence the deep reinforcement learning agent’s choices. The integration of DRL with explainable AI methods such as SHAP demonstrates significant potential for the advancement of robust and efficient counter-drone systems., Open Access funding provided thanks to the CRUE-CSIC agreement with Springer Nature. This work was funded by the Ministry of Science and Innovation of Spain under grant PID2020-116377RB- C21 for drone fleet management and optimization of commercial operations flight plans., Peer Reviewed, Postprint (published version)
Recercat. Dipósit de la Recerca de Catalunya
Publicación Contribución de congreso (ConferenceOutput)

Drone performance modelling for U-space strategic deconflicting services

  • Sánchez Segura, Albert
  • Pastor Llorens, Enric
  • Barrado Muxí, Cristina
Integrating drones into U-space airspace demands accurate performance modelling to support strategic deconfliction and reliable 4D trajectory predictions. However, most commercially available drones lack standardised performance profiles, making current ATM-oriented models like BADA not directly usable for U-space services. This paper presents a kinematic, operationally-focused approach to drone performance modelling based on classical dynamic principles and real flight data from representative drone platforms. The methodology divides flight into key phases, such as turns, climbs, and transitions, to generate polynomial estimators that capture essential dynamic behaviour without requiring complex sub-system modelling. The proposed models are packaged for easy integration by Uspace service developers and validated against actual drone flight trajectories. Results show improved prediction accuracy, highlighting the value of tailored drone performance models in enabling safe and efficient U-space operations., This research was supported by grants PID2023-152053OB-C21 and PID2020-116377RB-C21, funded by MICIU/AEI /10.13039/501100011033 and by FEDER, UE., Peer Reviewed, Postprint (published version)
Recercat. Dipósit de la Recerca de Catalunya
Publicación Artículo científico (JournalArticle)

Monitoring tomato leaf disease through convolutional neural networks

  • Guerrero Ibañez, Juan Antonio
  • Reyes Muñoz, María Angélica
Agriculture plays an essential role in Mexico’s economy. The agricultural sector has a 2.5% share of Mexico’s gross domestic product. Specifically, tomatoes have become the country’s most exported agricultural product. That is why there is an increasing need to improve crop yields. One of the elements that can considerably affect crop productivity is diseases caused by agents such as bacteria, fungi, and viruses. However, the process of disease identification can be costly and, in many cases, time-consuming. Deep learning techniques have begun to be applied in the process of plant disease identification with promising results. In this paper, we propose a model based on convolutional neural networks to identify and classify tomato leaf diseases using a public dataset and complementing it with other photographs taken in the fields of the country. To avoid overfitting, generative adversarial networks were used to generate samples with the same characteristics as the training data. The results show that the proposed model achieves a high performance in the process of detection and classification of diseases in tomato leaves: the accuracy achieved is greater than 99% in both the training dataset and the test dataset., This work was partially funded by the State Research Agency of Spain under grant number PID2020-116377RB-C21., Peer Reviewed, Postprint (published version)