Motion control of unmanned ground vehicle using artificial intelligence

The aim of this thesis is to solve two problems: the. trajectory tracking and navigation, for controlling the motion of unmanned ground vehicles (UGV). Such vehicles are usually used in industry for assisting automated production process or delivery services to improve and enhance the quality and efficiency. With regard to the trajectory tracking problem, the main task is to design a new method that is capable of minimising trajectory-tracking errors in UGV. To achieve this, a comprehensive mathematical model needs to be established that contains kinematic and dynamic characteristics beside actuators. In addition, different trajectories need to be generated and applied individually as a reference input, i.e. continuous gradient trajectories such as linear, circular and lemniscuses or a non-continuous gradient trajectory such as a square trajectory. The design method is based on a novel fractional order proportional integral derivative (FOPID) control strategy, which is proposed to control the movement of UGV to track given trajectories. Two FOPID controllers are required in this design. The first FOPID is constructed in order to control the orientation of UGV. The second FOPID controller is to control the speed of UGV. The particle swarm optimization (PSO) algorithm is used to obtain the optimal parameters for both controllers. The significance of the proposed method is that an observable improvement has been achieved in terms of minimising trajectory-tracking errors and reducing control efforts, especially in continuous gradient trajectories. The stability of the proposed controllers is investigated based upon Nyquist stability criterion. Moreover, the robustness of the controllers is examined in the presence of disturbances to demonstrate the effectiveness of the controllers under certain harsh conditions. The influence from external disturbances has been represented by square pulses and sinusoidal waves. The drawback of this method, however, a highly trajectory tracking error is observed in non-continuous gradient trajectories due to the sharpness of the rotation at the corners of a square trajectory. To overcome this drawback, a new controller, abbreviated as (NN-FOPID), has been proposed based on a combination of neural networks and the FOPID. The purpose is to minimise the trajectory tracking error of non-continuous trajectories, in particular. The Levenberg-Marquardt (LM) algorithm is used to train the NN-FOPID controller. The neural networks’ cognitive capacities have made the system adaptable to respond effectively to the variants in trajectories. The obtained results by using NN-FOPID have shown a significant improvement of reducing errors of trajectory tracking and increasing control efforts over the results by FOPID. The other task is to solve the navigation problem of UGV in static and dynamic environments. This can be conducted by firstly constructing workspace environments that contain multiple dynamic and static obstacles. The dynamic obstructing obstacles can move in different velocities. The static obstacles can be randomly positioned in the workspace and all obstacles are allowed to have different sizes and shapes. Secondly, a UGV can be placed in any initial posture on the condition that it has to reach a given destination within the boundaries of the workspace. Thirdly, a method based on fuzzy inference systems (FIS) is proposed to control the motion of the UGV. The design of FIS is based on fuzzification, inference engine and defuzzification processes. The navigation task is divided into obstacle avoidance and target reaching tasks. Consequently, two individual FIS controllers are required to drive the actuators of the UGV, one is to avoid obstacles and the other is to reach a target. Both FIS controllers are combined through a switching mechanism to select the obstacle avoidance FIS controller if there is an obstacle, otherwise choosing reaching target FIS. The simulation results have confirmed the effectiveness of the proposed design in terms of obtaining optimal paths with shortest elapsed time. Similarly, a new method is proposed based on an adaptive neurofuzzy inference system (ANFIS) to guide the UGV in unstructured environments. This method combines the advantages of adaptive leaning and inference fuzzy system. The simulation results have demonstrated adequate achievements in terms of obtaining shortest and feasible paths whilst avoiding static obstructing obstacles and hence reaching the specified targets speedily. Finally, a UGV is constructed to investigate the overall performance of the proposed FIS controllers practically. The architecture of the UGV consists of three ultrasonic sensors, a magnetic compass and two quadratic decoders that they are interfaced with an Arduino microcontroller to read the sensory information. The Arduino, who acts as a slave microcontroller is serially connected with a master Raspberry Pi microcontroller. Raspberry Pi and Arduino communicate with each other based on a proposed hierarchical algorithm. Three case studies are introduced to demonstrate the effectiveness and the validation of the proposed FIS controllers and the UGV’s platform in real-time.