Deep Fuzzy Neural Network-Based Control

In this work, we presented an online learning method for improved control of nonlinear systems by combining deep learning and fuzzy logic. Given the ability of deep learning to generalise knowledge from training samples, the proposed method requires minimum amount of information about the system to be controlled. However, in aerial robotics, where the operating conditions may vary, online learning is required. In this study, fuzzy logic is preferred to provide supervising feedback to the deep model for adapting to variations in the system dynamics as well as new operational conditions. The learning method is divided into two phases: offline pre-training and online post-training. In the former, the system is controlled by a conventional controller and a deep fuzzy neural network (DFNN) is pre-trained based on the recorded input-output dataset, in order to approximate the inverse dynamical model of the system. In the latter, only the pre-trained DFNN is used to control the system. In this phase, the fuzzy logic, which encodes the expert knowledge, is utilized to observe the behaviour of the system and to correct the action of DFNN instantaneously. The experimental results show that the proposed online learning-based approach improves the trajectory tracking performance of the unmanned aerial vehicle. (paper1, paper2)

Fuzzy Mapping-Based Control

Although a considerable amount of effort has been put in to show that fuzzy logic controllers have exceptional capabilities of dealing with uncertainty, there are still noteworthy concerns, e.g., the design of fuzzy logic controllers is an arduous task due to the lack of closed-form input-output relationships which is a limitation to interpretability of these controllers. The role of design parameters in fuzzy logic controllers, such as position, shape, and height of membership functions, is not straightforward. Motivated by the fact that the availability of an interpretable relationship from input to output will simplify the design procedure of fuzzy logic controllers, the main aims in this work are derive fuzzy mappings for both type-1 and interval type-2 fuzzy logic controllers, analyse them, and eventually benefit from such a nonlinear mapping to design fuzzy logic controllers. Thereafter, simulation and real-time experimental results support the presented theoretical findings. (paper1, paper2, paper3)

Artificial Neural Networks-Based Control

In this work, a learning model-free control method is proposed for accurate trajectory tracking and safe landing of unmanned aerial vehicles (UAVs). A realistic scenario is considered where the UAV commutes between stations at high-speeds, experiences a single motor failure while surveying an area, and thus requires to land safely at a designated secure location. The proposed challenge is viewed solely as a control problem. A hybrid control architecture – an artificial neural network (ANN)-assisted proportional-derivative controller – is able to learn the system dynamics online and compensate for the error generated during different phases of the considered scenario: fast and agile flight, motor failure, and safe landing. The tuning of weights is not required as the structure of the ANN controller starts to learn online, each time it is initialised, even when the scenario changes – thus, making it completely model-free. Moreover, the simplicity of the neural network-based controller allows for the implementation on a low-cost low-power onboard computer. Overall, the real-time experiments show that the proposed controller outperforms the conventional controller. (paper1, paper2)

Fuzzy Logic-Based Control

Input uncertainty, e.g., noise on the on-board camera and inertial measurement unit, in vision-based control of unmanned aerial vehicles (UAVs) is an inevitable problem. In order to handle input uncertainties, as well as further analyze the interaction between the input and the antecedent fuzzy sets of non-singleton fuzzy logic controllers (NSFLCs), an input uncertainty sensitivity enhanced NSFLC has been developed. Based on recent advances in non-singleton inference, the centroid of the intersection of the input and antecedent FSs (Cen-NSFLC) is utilized to calculate the firing strength of each rule instead of the maximum of the intersection used in traditional NSFLC (Tra-NSFLC). An accurate monocular keyframe-based visual-inertial simultaneous localization and mapping (SLAM) approach is used to estimate the position of the quadrotor UAV in GPS-denied unknown environments. All controllers are evaluated for different flight speeds, thus introducing different levels of uncertainty into the control problem. Visual-inertial SLAM-based real time quadrotor UAV flight tests demonstrate that not only does the Cen-NSFLC achieve the best control performance among the four controllers, but it also shows better control performance when compared to their singleton counterparts. (paper1, paper2, paper3, paper4)