Artificial neural network-based positioning error modeling and compensation for low-cost encoders of four-wheeled vehicles.

Several academic and budgetary applications in robotics use low-cost encoders which usually present errors inherent to the fabrication of their components or the surrounding environment. However, the data gathered from these sensors could be used successfully if the estimations based on the data are be compensated. This note presents an efficient method to compensate the error of estimation for the distance traveled by robotic four-wheeled vehicles with a speed control based on pulse width modulation. The current approach uses a methodology based on artificial neural networks which compensate the estimation error and enables a way to model the error for a further method of position estimation. Precisely, the model proposed in this work is based on a neural network with one hidden layer and five nodes. Our approach uses the information from the pulse width modulation value to control the speed of the vehicle as well as the values for the tick counting of the encoders in the four wheels. The back-propagation algorithm was used to calculate the weights of the nodes in the neural network. This method showed an improvement in the results with an error comparable to the intrinsic error due to the precision of the sensor, and gives an error model based on a zero-mean normal distribution. [ABSTRACT FROM AUTHOR]