Neural Adaptive Flight Control Testing on an Unmanned Experimental Aerial Vehicle

Abstract

Unmanned Aerial Vehicles have demonstrated potential as being effective platforms for supporting scientific and exploratory missions. They are capable of performing long endurance flights, and reaching remote areas that may be too dangerous for humans. As their role and types of missions expand, challenges are presented which require onboard systems to have increasingly higher levels of intelligence and adaptability. Missions requiring radical reconfiguration to carry mission-specific payloads, or operations under uncertain or unknown flight conditions, will require intelligent flight controllers that are capable of being deployed with minimal prior testing. This paper describes the testing of a neural adaptive flight controller that was designed to provide consistent handling qualities across flight conditions and for different aircraft configurations. The controller was flight tested on an unmanned experimental aerial vehicle, without the benefit of extensive gain tuning or explicit knowledge of the aircraft’s aerodynamic characteristics. An overview of the neural adaptive flight controller is presented, along with a description of the experimental aerial vehicle test platform, and flight test results that demonstrate a dramatic improvement in handling qualities resulting from neural adaptation. I. Introduction HE Intelligent Flight Control (IFC) project at NASA Ames Research Center endeavors to investigate, maturate, and validate the next generation of intelligent flight controllers that exhibit higher levels of adaptability and autonomy than the current state of the art, reduce the costs associated with flight control law development, and can be applied to a wider-range of vehicle classes without significant development costs. The current IFC architecture is based on neural network augmentation, and is designed to enhance the handling qualities and response of a vehicle system subject to control surface failures or uncertainty in vehicle aerodynamic response resulting from structural damage, failures, or model inaccuracy. This technology has the promise to increase overall vehicle safety by adapting to changes in aircraft dynamics due to damage or failures, reduce cost associated with flight control law development by providing consistent handling qualities across flight regimes and variable aircraft configurations, and allow the application of generic control designs over a wide-range of vehicle classes, for example from commercial transports to high performance military aircraft and experimental concepts. The process of validating experimental control technologies typically progresses from analytical analysis through testing using increasing levels of simulation fidelity to full-scale vehicle testing. Simulation testing has taken on increased importance over the past few decades. The rapid increase in computational power and tool sophistication available to researchers has allowed for more comprehensive testing and validation to be performed in the lab environment while providing results in a much timelier fashion. Analysis tools such as Matlab and Simulink integrate analysis tools with simulation capability seamlessly, and provide mechanisms for converting these designs directly to source code that can be quickly integrated into embedded flight vehicle control systems. Despite the dramatic advances in computational technology, a crucial step in the maturation and validation of any research control technologies is real-world experimental flight testing on fully developed aircraft systems. The magnitude and severity of implementation-specific artifacts on a theoretical control construct may not be fully