This project addresses the design and implementation of automatic control systems for aircraft with large flight envelopes, using a novel system identification approach based on nonparametric machine learning techniques. Conventional parametric models often poorly characterize the aerodynamics of autonomous aircraft, operating over a large range of speeds, power settings, and angles of attack. Although linearized aerodynamic models have been used for aircraft stability analysis and flight controls design for many decades, the increasing interest in autonomous aircraft with expended flight envelopes limits the use of this approach. Larger angle of attack and side-slip ranges allow flight with more extreme wind conditions and with increases maneuverability required for unmanned aircraft.
Recent progress in machine learning and big-data systems allow more complete and robust models suitable for use in aircraft and control system design. Our previous work suggests that the application of these learning techniques, leading to nonparametric aerodynamic models, along with existing or enhanced sensing systems, can be used to control aircraft over a range of maneuvers and flight conditions that would not be possible based on simple parametrized small aircraft configurations. The aerodynamic modeling and control system design techniques proposed here have been shown to be applicable to spin detection and recovery, but will likely be useful in the execution of other maneuvers important to pilotless aircraft, such as high angle of attack maneuvering, vertical landing, and perching. These maneuvers offer the potential to extend the existing capabilities of autonomous aircraft, allowing them to operate in more constrained environments.