A Comparative Study of Nonlinear Control and Passivity-Based Control using Neural Networks for A Bicycle Robot

In this paper, a comparative study of nonlinear control and passivity-based control using neural networks for a bicycle robot is proposed. Bicycle robot is a nonlinear, multi-input multi-output system. Two inputs of a bicycle robot are the steering torque and kinetic energy. Its two outputs are the steering angle and the rolling angle. The control problem is that the steering angle and the rolling angle track a value of zero, and the velocity of the steering angle and velocity of the rolling angle track a value of zero to make a bicycle robot stabilize at its vertical balance. Firstly, an input-output linearization control law decouples the bicycle robot into single-input single-output systems. This plant is passive and zero-state observable. Secondly, the passivity-based control law is applied to each single-input single-output system. Finally, the neural network, which performs the passivity-based control, is applied to each single-input single-output system in order that the bicycle robot keeps its vertical balance. A training algorithm using the steepest descend method is proposed. The simulation results of the passivity-based control and the results of the passivity-based control using neural networks show that the bicycle robot keeps its vertical balance. The settling time of the steering angle and the rolling angle of the passivity-based control using a neural network, 1.8s, is shorter than that of the passivity-based control. There is a comparison with the passivity-based control combined with sliding mode control for a bicycle robot.