Towards Multi-Aircraft Transfer Learning for Trajectory Tracking

Abstract

In this paper, a control method that combines model reference adaptive control (MRAC) and iterative learning con- trol (ILC) is applied to aircraft trajectory tracking. ILC is intended for repetition-invariant system dynamics, in which the nominal dynamic model of the aircraft is assumed to be known. However, in real operations, this requirement is not met, as different aircraft perform consecutive flights along the same trajectory. To address this limitation, a multi-aircraft transfer learning strategy is proposed, which allows the learned trajectory knowledge to be transferred to dynamically different aircraft at each iteration. In order to achieve this, the aircraft’s baseline controller is augmented with an MRAC, which drives the system’s performance close to that of a reference model, and an ILC, which serves as a high-level adaptation scheme to compensate for repetitive disturbances. According to the preliminary results obtained from experiments carried out with various simulated aircraft, taking into account model uncer- tainties, disturbances, and changing dynamics, the performance of the ILC in combination with the MRAC augmentation of a baseline controller is superior to that of the baseline controller without MRAC augmentation. After a few iterations, a signifi- cant reduction in the trajectory tracking error is observed, with only small fluctuations occurring throughout the subsequent iterations. As a result, the MRAC- ILC combination makes the ILC applicable to real operations, improves the predictability of aircraft trajectories, and enhances the efficiency of the air traffic management system.