Torque control design of nonholonomic mobile robots using a neural network-based approach

In this thesis, a novel control algorithm is proposed for a nonholonomic mobile robot with completely unknown robot dynamics and subject to bounded unknown disturbances. By taking advantage of the robot regressor dynamics, the neural network assumes a single layer structure. The learning algorithm is derived from Lyapunov stable analysis, which is much simpler than most commonly used neural network learning algorithms. The control algorithm is computationally efficient resulting from the simple neural network structure and its simple learning rule. The proposed controller is capable of achieving precise motion control of a nonholonomic mobile robot through the on-line learning ability. The stability of the proposed controller is proved using a Lyapunov stability theory.; In addition, the proposed controller is extended to a mobile robot with unknown kinematic parameters, where the linear and angular velocities are chosen as the velocity control input. The extended controller is capable of dealing with completely unknown kinematics and dynamics parameters of the robot system. One neural network is designed to learn both the dynamics and kinematic parameters.; Furthermore, a novel torque controller is proposed for nonholonomic mobile robots with obstacle avoidance. By introducing an obstacle torque in the controller, the proposed controller is capable of driving the robot to its target and avoiding obstacles in various environments. Both the neural network structure and its learning algorithm are simple. The controller is guaranteed to be stable and converge by a Lyapunov stability theory, subject to unmodeled unstructured disturbance. No prior information of the environment is needed, all the environment information needed can be obtained from the on-board robot sensors with a limited visibility range.