Bio-Inspired Adaptive Fuzzy Control Systems for Precise Low-Altitude Hovering of an Unmanned Aerial Vehicle Under Large Uncertainties

The ability to learn and adapt to unknown system dynamics simplifies controller design and enables complex platforms to be controlled without the need to build complex mathematical models. Taking some inspiration from the way humans learn, we present the concept of bio-inspired self-learning in aerial robotics, leveraging on the concept of an adaptive Takagi-Sugeno (TS)-fuzzy control system. The main distinguishing feature of the evolving TS-fuzzy system is the ability to learn from scratch, eliminating the need to have a-priori knowledge about the system as in the traditional model-based control systems. Besides, the system can also learn from certain predefined rules. As opposed to traditional fuzzy systems, which require prior training (knowledge) to build their structure, the evolving TS-fuzzy system needs no such prior knowledge since the controller can perform online self-learning. Also, its ability to capture high-degree of uncertainties (e.g. severe ground effects due to low-altitude flying) is very advantageous. To demonstrate the efficacy of the control systems, we design and implement the evolving TS-fuzzy autopilots in the five control loops of our Tarot hexacopter drone after conducting extensive computer simulations using non-linear aerodynamics models. We also compare the efficacy of the autopilot systems with respect to the effectiveness of traditional PID controllers in the altitude control loop as a benchmark.