Modelling Self-Balancing Robot using ARX and Neural Network

This paper considers how to model a two-wheeled self-balancing robot (TWSB) using ARX algorithm and artificial neural network to improve the efficiency of PID control system. The dynamic model of the robot is developed using state-space representation, and system identification is performed to compare linear models such as ARX and other models including neural networks. The implementation of the neural network algorithm demonstrates improved stability and faster response time, achieving balance in less than 5 seconds. Experimental results show a significant reduction in error, with the Mean Square Error (MSE) reaching as low as 0.082256, highlighting the significant differences between the two-wheeled self-balancing robot and the two-wheeled self-balancing robot.