Automated Air Traffic Control
for Non-Towered Airports
For my Ph.D. thesis, I modeled the behavior of aircraft in the airport pattern as a hidden Markov Model (HMM) whose parameters are learned from real-world radar observations. Then I determined optimal advisories to reduce the risk of collision by formulating the problem as a partially observable semi-Markov decision process (POSMDP), and solved it using Reinforcement Learning techniques.
In order to address the computational complexity of solving the problem, I used different approximation methods including exponential sojourn times, phase-type distributions, online algorithms, and particle filters for belief estimation.