Master Thesis: Triage of Non-compliant UAS Flights Using Machine Learning Methods

Master Thesis:
Triage of Non-compliant UAS Flights Using Machine Learning Methods
Job description   Background
With the steep increase in amateur drone flights in urban centres around the world, authorities are searching for effective ways to maintain an orderly and safe traffic situation. In the Netherlands, due to the popularity of some drone-flying spots and the high density of urban centres, many drone flights take place in restricted areas or close to high-importance buildings. For example, the city centre of Amsterdam often has days with hundreds of unique drone flights per day, most of which are lacking a permit to fly within the Schiphol Airport CTR.
However, due to the high number of flights, as well as the difficulty to accurately determine the location of pilots, the local authorities are unable to prevent, pursue, and investigate them all of the ones that breach the rules. Many such flights occur due to the lack of awareness from pilots, and thus pose a low threat. However, these might obscure other flights that can produce harm and are of a high threat level.
The goal of this thesis is to develop an online system that, based on live drone detection data, can determine the threat level of a drone flight and inform authorities whether it must be immediately addressed. One of the current limitations of such methods is the need for annotating a large number of drone trajectories. Reinforcement learning or equivalent methods can generate artificial trajectories that eliminate the need for manual annotation.

Thus, artificial drone flights can be created that cover a wide range of threat levels, leading to a better coverage than real data. These can be labelled explicitly, or implicitly through the use of clustering or unsupervised learning methods.
Tasks
The assignment will include the following tasks:
Investigation of drone traffic in the Netherlands and existing approaches for this problem;
Implementation of an appropriate machine learning environment using the Blue Sky-Gym library;
Model selection, tuning or development (with algorithms such as SAC from stable-baselines3 or other);
Implementation of a classification method for real drone trajectories based on the artificial trajectory dataset;
A robustness analysis and benchmark of the accuracy of the developed model, and future work that must be performed to reach desirable results.
Results
The final outcome of this assignment will be:
An ML-based method and/or workflow that can determine the threat level of a drone flight;
A technical thesis report describing the approach, results and conclusions of the work;
Optional, but encouraged: a conference and/or journal publication.
Duration
Standard duration of a TU Delft Aerospace Engineering MSc thesis.
What do we expect from you?
You are an MSc in Aerospace Engineering.
You have experience with programming in Python.
You have experience with practical application of ML/RL (PyTorch, or others).
You have completed a machine learning course (e.g., DSAIT
4005, DSAIT
4115) and preferably the Air Traffic Management course (AE4321-15).
What we offer
Enthusiastic colleagues who are experts in their field;
A flexible working space;
An environment where you have the opportunity to develop your skills and learn new ones;
A challenging assignment in a high-tech, result orientated work environment;
A thesis assignment remuneration;
An informal corporate culture where your opinion counts!
About NLR
You will be working within the Air Traffic Management & Airports department. Your colleagues are focused on solving real-world problems within air traffic management, airspace design, U-Space and other exciting domains.

#J-18808-Ljbffr