Collision risk assessment in terminal manoeuvring areas based on trajectory generation methods

Collision risk models are the mainstay of collision avoidance studies. We address here methods based on Monte Carlo simulations to estimate collision risk probabilities. As observations of real trajectories are not enough to estimate probabilities of losses of separation, which are rare events, generated synthetic trajectories address this limitation. In this paper, we build upon a recently published data-based statistical trajectory generation method based on Variational Autoencoders and estimate the risk of mid-air collision between arriving aircraft that are in the downwind and departing aircraft that are climbing-out in a southerly direction at Paris–Orly airport. We compute the probabilities of losses of separation and discuss the difference by an order of magnitude that we found in the results, depending on the runway configuration in place