Dynamic Optimization of Target Search by a Fleet of Drones Using Artificial Intelligence Techniques

Job description:

Introduction / context:

The PhD position is offered by the DRIVE laboratory of the Université Bourgogne Europe, located in Nevers, France. Staffed by ca. 60 community members of which ca. 30 researchers and ca. 20 PhD. students, the DRIVE laboratory develops high-level applied and fundamental research with cutting-edge equipment. This PhD project is funded by a ministerial grant and focuses on target search.

The optimization of target search is a problem that first emerged during World War II when American researchers worked on the optimal detection of enemy submarines using a fleet of aircraft. Over the years, numerous new applications have arisen, demonstrating the tangible benefits of optimization methods. Some notable examples include: (i) searching for shipwrecks at sea, such as the USS Scorpion, the SS Central America, or the black boxes of Air France Flight AF447; (ii) early fire detection; (iii) search and rescue operations, particularly following disasters like Hurricane Katrina or when locating missing hikers with rescue teams; (iv) tracking and apprehending fugitives, especially in difficult-to-access areas such as mountains and forests. Most of these applications rely on the use of drones, whether aerial or underwater. With recent technological advances in drones, interest in this field has continued to grow, leading to an increasing number of studies on the subject. The fundamental problem underlying most of these studies is the optimization of search time distribution to maximize the probability of target detection, which translates into a nonlinear optimization problem. Despite extensive research on this topic, the predominant approach remains offline, meaning it relies on pre-planned strategies without allowing drones to adapt in real-time to newly collected information. This PhD project specifically addresses this issue by developing optimization techniques and/or artificial intelligence methods to dynamically organize target search, adjusting to updated data during the mission. To achieve this, it is crucial to maintain continuous connectivity between each drone and the base station, ensuring that the collected data can be processed and that new instructions can be transmitted accordingly. However, in many applications, network quality is insufficient, leading to unstable connections. It is therefore necessary to consider a multi-hop network, where some drones, out of direct range of the base station, use other drones as relays to ensure data transmission.

Scientific context:

The PhD project aims to develop technological solutions that enable drones to maintain continuous connectivity with the base station, relying on relay drones while dynamically optimizing the probability of target detection. To achieve this objective, several scientific challenges will be addressed: (i) Connectivity: Modeling connectivity in the context of search time distribution to maximize the probability of detection. Although connectivity has been studied in the past for target search, it has always been approached in highly simplified forms (e.g., patterns) and without considering resource distribution, which is crucial in most real-world applications. The main challenge lies in the dual role of drones, which must both search for the target and act as relays to transmit information to other drones. Connectivity quality directly depends on the distance between drones and the base station, making its modeling particularly complex. (ii) Dynamic decision-making: Modeling a dynamic search problem that can adapt resource allocation in real-time based on newly collected information, along with the development of AI-based or OR-based solution methods. As with connectivity, existing dynamic approaches do not account for resource distribution optimization, representing a major gap in this field. (iii) Gamification: Transforming the dynamic optimization problem—intrinsically complex and challenging to solve—into a game where an AI can learn and perform efficiently. The advantage of this approach is that it offers a flexible method that can be easily adapted to various problem variations without requiring significant research and development efforts for each model modification. Although gamification is a promising approach, it has never been applied to the dynamic target search problem before.

Work envisaged:

The candidate will begin by conducting a literature review to study the various existing approaches for modeling connectivity in the context of target search. Then, the problem of continuous connectivity in this context will be analyzed and formalized. A modeling approach based on a dynamic graph and a flow problem will be specifically explored. The next step will involve developing artificial intelligence or optimization methods to dynamically adapt target search to newly collected information and/or connectivity constraints. Finally, a gamification approach will be explored: this will involve transforming the dynamic optimization problem into a game, allowing generalist AIs to learn and perform efficiently.

The provisional schedule is as follows:

• Phase 1: Bibliographic study.
• Phase 2: Removal of scientific barriers and development of operational strategies adapted to the system (localization, planning and cyber security).
• Phase 3: Testing and validation of the use cases by simulation as well as the implementation of a proof of concept with drones.
• Phase 4: Writing of the thesis.

The Ph.D. position is proposed by the DRIVE lab of the Universityé Bourgogne Europe, located in Nevers Magny-Cours in France. Staffed by ca. 60 community members of which ca. 30 researchers and ca. 20 PhD. students, the DRIVE laboratory develops high-level applied and fundamental research with cutting-edge equipment. The research work encompasses 2 areas of specialism: intelligent systems with energy optimization as well as mechanics of materials and structures.

Candidates should have a Master's degree or an engineering degree in computer science or telecommunications. A good grounding in Operations research, optimization and Artificial Intelligence as well as practical skills in programming and software tools (e.g. Python, C++, Cplex) and fluent English (written and spoken) are required. Candidates must be motivated to learn quickly and work effectively on challenging research problems.