A rational agent framework for adaptive flight control of UAVs

This paper presents a new intelligent agent framework that combines rational decision making with abstractions of sensing and actions for unmanned aerial vehicle (UAV) control systems. The objective of the new framework is to provide self tuning and more adaptability to an onboard autopilot. This is achieved through controlloer switching and tuning by the agent in face of dynamic changes affecting the plane due to aerodynamic and onboard system changes. The framework integrates onboard dynamical system identification, hybrid adaptive control, model validation and diagnosis. The agent, tuning and supervising the autopilot, also operates real-time evaluations of abstractions on the effects of its actions in the future. The methods are demonstrated in simulation on a benchmark performance of an Aerosonde UAV.