Sequential Monte Carlo Optimisation for Air Traffic Management

TL;DR

This paper presents a real-time Monte Carlo optimization algorithm within a Model Predictive Control framework for air traffic management, achieving significant fuel savings and trajectory optimization in complex, stochastic scenarios.

Contribution

It introduces a GPU-accelerated Monte Carlo method for real-time, multi-aircraft trajectory optimization in the TMA, incorporating stochastic wind models and safety constraints.

Findings

Achieved approximately 30% fuel savings in simulations.

Demonstrated real-time performance on GPU with complex scenarios.

Showed flexibility by integrating noise-reduction objectives.

Abstract

This report shows that significant reduction in fuel use could be achieved by the adoption of `free flight' type of trajectories in the Terminal Manoeuvring Area (TMA) of an airport, under the control of an algorithm which optimises the trajectories of all the aircraft within the TMA simultaneously while maintaining safe separation. We propose the real-time use of Monte Carlo optimisation in the framework of Model Predictive Control (MPC) as the trajectory planning algorithm. Implementation on a Graphical Processor Unit (GPU) allows the exploitation of the parallelism inherent in Monte Carlo methods, which results in solution speeds high enough to allow real-time use. We demonstrate the solution of very complicated scenarios with both arrival and departure aircraft, in three dimensions, in the presence of a stochastic wind model and non-convex safe-separation constraints. We evaluate our algorithm on flight data obtained in the London Gatwick Airport TMA, and show that fuel saving of about 30% can be obtained. We also demonstrate the flexibility of our approach by adding noise-reduction objectives to the problem and observing the resulting modifications to arrival and departure trajectories.