Advancing Fault-Tolerant Learning-Oriented Control for Unmanned Aerial Systems

The rapid advancement of automatic control technology has sparked significant interest among researchers in creating more reliable and simplified models of unmanned aerial vehicles (UAVs). This interest is motivated by the need to enhance the performance and resilience of these systems in challenging conditions, such as wind gusts and adverse weather. This paper presents novel strategies for enhancing the resilience of unmanned aerial systems (UAS) with fault-tolerant control (FTC) by learning-oriented control and a constructive fault estimation with Proportional-Integral (PI) observer. The learning-control is deep-deterministic policy gradient (DDPG) which is trained in only one state but used beyond its environment for other states to control. The faults are designed in three divergent conditions and the augmented PI observer is responsible in capturing them. The success of estimating the faults is used for this FTC to compensate the faulty system with learning-oriented control as the advancement of the FTC. The proposed approach has the potential to enhance the performance and resilience of UAVs, thus contributing to the development of more robust and reliable systems.