Artificial Intelligence Applications for Drones Navigation in GPS-denied or degraded Environments

Department: PhD in Aerospace Engineering
Founded by: Leonardo Company
Supervisor: Prof. Giorgio Guglieri | Prof. Fabio Dovis
Candidate: Simone Godio

One of the strongest limits in robotic applications is not being able to understand where your system is located autonomously with satisfying precision. Therefore, it becomes difficult to guide the robot’s movement. To overcome this problem, it is usually possible to use external aids such as GNSS, Motion Capture Cameras, Total Stations, or similar. Also, Ultra-Wideband technologies are recently taking old as a cheaper source to localize the vehicle. However, these systems cannot be adopted in critical (GPS-denied) and unknown areas without external systems. Therefore, many of the robotic applications are limited in these scenarios.
That’s why recently several open-source algorithms (ORB-SLAM, SVO, VINS, Okvis, Rovio, and many others) have been introduced to perform Visual Inertial Odometry for complete autonomous applications. This particular technique performs a sensor-fusion between optical sensors as mono or stereo cameras and inertial sensors to estimate the trajectory elaborated from an initial position.
Besides, research is still truly open to innovative lighter, and more robust solutions to this problem on board the aircraft itself, and this is one of the goals of our research group.

We are also investigating different approaches to solve a coverage planning problem for a fleet of Unmanned Aerial Vehicles exploring critical areas.
The main goal is to fully cover the map, maintaining a uniform distribution of the fleet on the map, and avoiding collisions between vehicles and other obstacles. This specific task is suitable for surveillance applications, where the uniform distribution of the fleet in the map permits them to reach any position on the map as fast as possible in emergency scenarios.
To solve this problem, we are proposing more neural network techniques to optimize the coverage time and the homogenous distribution of the feet in the environment.

  • Improvement of innovative AI-based autonomous localization techniques through RGB, and depth images, and Inertial Sensors Fusion.
  • Implementation of innovative algorithms to manage a fleet of collaborative drones in critical GPS denied or degraded environments.
  • Development of control low ad hoc for autonomous navigation in these types of environments.