Predictive sectorization and bayesian optimized consensus for admission control in autonomous airspace operations

Abstract

Modern air traffic management distributes authority across many controllers and facilities, but within each sector a single controller is the sole authority for separation and sequencing. As traffic volumes grow and next generation concepts such as NASA’s Advanced Air Mobility push toward increasingly autonomous flight, this per sector human dependency becomes a scaling bottleneck. This thesis explores a multi stage pipeline for adaptive airspace sectorization and decentralizedadmission control that can reduce per sector workload without removing human oversight. Partitioning airspace into sectors is not straightforward. The grid size affects workload, handoff frequency, and the capacity of whatever coordination mechanism operates within each sector. The right partition depends on traffic density, flow direction, and altitude distribution, and the consensus protocol itself introduces tunable parameters that interact with the sectorization choice. Finding good configurations for both requires a data driven approach. We begin by simulating replays of real FAA SWIM flight plan trajectories across a three dimensional discretized airspace grid. By scoring every valid grid partition on safety, efficiency, and density balance criteria, the simulator produces a labeled dataset mapping traffic features to optimal sectorization grids. A two stage XGBoost predictor trained on this dataset first determines whether sectorization is needed, then states the optimal grid configuration. The predicted grid defines sectorboundaries that are handed to a Paxos based consensus protocol, in which the aircraft coordinate sector transitions. Because the protocol exposes several tunable parameters, a Bayesian optimization loop driven by a Gaussian Process surrogate, searches for configurations that maximize entry success rate while rejecting any setting that produces a near mid air collision. Results show that the predictor reliably selects appropriate grid configurations across diverse traffic conditions, the consensus protocol maintains high entry success rates as traffic scales from light to heavy demand, and Bayesian Optimization discovers that each operating environment requires a qualitatively different protocol tuning, confirming that no single default setting is sufficient. Together, these stages demonstrate that machine learned sectorization combined with consensus basedadmission control can adapt to varying traffic conditions while maintaining separation safety, offering a viable path toward autonomous airspace operations.