Evolving and Certifiable Autopilot for Unmanned Aerial Systems

The project consist of the development of a new intelligent flight control system with learning capabilities and a high degree of assurance, that can be certified by the FAA Machine learning and artificial intelligent research has led to many tangible results and recent developments in cognitive control and decision making. Although automatic flight controllers are widely used and they have become common in recent years, they often lack intelligence, adaptability, and high performance. Reliability of UASs in unforeseen conditions is a direct function of their intelligence and adaptability. The proposed project aims to take advantage of high-performance computing platforms and the state-of-the art machine learning and verification algorithms to develop a new intelligent, adaptable, and certifiable flight control system with learning capabilities. The autopilot system will be able to learn from each flight experience and develop intuition to adapt to a high level of uncertainties. To provide a high degree of assurance and to make the learning autopilot system safe and certifiable, a secondary and conventional autopilot system will be integrated based on the run-time assurance architecture. A monitor will be developed to continuously check aircraft states and envelope protection limits, and handover aircraft control to the conventional autopilot system if needed. Provable guarantees of the monitor and the controllers will be provided using formal analysis. The propose a hybrid flight control system which has adaptability and intelligence of skilled pilots and at the same is cable of performing complex analysis and decision making algorithms in real-time. We aim to build and train an artificial neural network model that can mimic the performance of the classical robust optimal controllers, extend the robustness, adaptability, and curiosity of the artificial neural network controller and integrate a Real-Time Assurance (RTA) system. More »