**Problem** Modern digital fly-by-wire control systems are designed around the aerodynamics and mass properties of each aircraft. Such systems are not robust to variations of those properties (e.g., failure mode or mass distribution of flexible wings). When faced with large deviations, the performance of the flight control system degrades significantly, necessitating human intervention and impacting safety. This is a barrier to both modular/flexible next-gen designs and path-to-autonomy. **Solution** Artificial intelligence (AI) systems have a role to play in making flight control systems safer, more robust and cost effective. The project will use low-cost drone platforms as a cost-effective and risk-managed stepping stone towards complex civil flight operations. Through a combination of simulated and targeted physical flight tests we will develop an adaptive neural flight controller for forward-flight and vertical take-off and landing (VTOL) configurations, with hardware validation focused on fixed-wing forward flight and VTOL hover, and simulation-based validation used for more complex hybrid forward-flight and transition-flight regimes, and demonstrate how adaptive AI opens up unprecedented design possibilities for next-gen aviation. **Partner Expertise** * Luffy AI is an SME developing ultra-robust, neuroplastic control networks capable of dynamically adapting to changes and disturbances. * The University of Southampton has a world-leading aerospace engineering research group and has been researching AI techniques for flight control, in particular domain randomisation techniques for Deep RL algorithms. **Outcome** The research contributions from both sides will be integrated to produce a neuroplastic AI flight controller trained with state-of-the-art domain randomisation. Model aircraft will be physically flown, where appropriate, to confirm flight performance, complemented by high-fidelity simulation for flight regimes that are not validated in hardware within this project. The AI flight control technology resulting from this project will provide benefits in terms of development efficiency, fault tolerance and robustness to turbulence in current aircraft designs, as well as being a key enabler for the stability and control of next-gen aircraft, including advanced air mobility vehicles, modular bodies, highly flexible/morphing wings and distributed actuation.

Luffy AI is developing a novel Adaptive Neural Network systems targeted at embedded robotic and industrial control. Our approach overcomes the "reality gap" that limits the application of the current generation of neural network systems in these sectors. In order to achieve wide acceptance within process industries, the developed technology must be proved safe, and a large number of stakeholders must understand how the AI technology impacts them and the process that is being controlled. The challenge is to cover needs of all stakeholders from the human operating the process to board room decision makers. In this grant project we undertake a feasibility study of applying Adaptive Neural Network controller into different processes used in the Foundation Industries. As an output, the project delivers releases a white paper discussing the potential applications of neural network controllers within the foundation industries, and the steps required to ensure trusted and responsible deployment. The project would culminate in formation of a consortium that would aim to deploy Adaptive Neural Network controller to a selected industrial process.

AI systems have a role to play in making flight control systems safer, more robust and cost-effective. Luffy AI has developed adaptive neural networks that allow real-time onboard learning and adaption. In this project, we will develop and demonstrate through simulation and flight testing an AI flight controller that can adapt to variations in both a drone flight vehicle and its environment. In addition, the project will develop a framework for the safety case for the regulation of AI based autonomy in aviation. This work led by the Institute for Safe Autonomy, at the University of York, will support the UK's wider regulatory ambitions in this area and position the UK as a leader in the safe deployment of autonomous systems for commercial use. Overall, this new capability will demonstrate how AI controllers can provide an alternative, cost-effective solutions for Urban Air Mobility (UAM) and large civil applications and improve their safety case in the eyes of regulators.

Luffy AI is developing a novel neural network system targeted at embedded robotic and industrial control. Our approach overcomes the "reality gap" that limits the application of the current generation of neural network systems in these sectors. Our first major application, and the focus of this grant project, is the development of a self-learning neural controller for robotic serial manipulators such as robotic arms. Our neural network controller will allow us to address the control issues currently limiting robotic manipulators from achieving the human like mobility and energy efficiency that will be essential for industrial applications. The grant project would culminate in the integration of the developed controller into a prototype robotic manipulator that can be demonstrated to prospective customers / industrial partners. Our controllers are also intended to be deployed on industrial control applications, where conventional control schemes provide suboptimal results.