A coordinated control for unmanned aircraft taxiing system via integrating reinforcement learning and adaptive strategy

Abstract Aircraft taxiing under a tractor reduces operating costs, improve airport operational efficiency, and decrease pollutant emissions. However, the unmanned aircraft–taxiing system is more complex and becomes more sensitive to system parameters and external uncertainties. The flexibility of motion is also constrained once the aircraft and tractor are formed into a whole. This study proposes a coordinated controller by integrating the advantages of model-based and learning-driven approaches. Firstly, a velocity control law based on virtual variable is designed by decoupling the mathematical model of the system. To enhance coordination adaptability, adaptation strategy is used to adjust gain parameters. And then, a reinforcement learning method is designed to handle system’s coupling and uncertainties. Basic motion policy and dynamic adaptation module ensure the learning ability of the agent. Finally, a set of simulations and experiments are carried out under different conditions, the results prove that the combined controller ensures coordinated motion under different conditions.