In this paper, we describe a robot endeffector tracking system based on two neural networks. The designed networks are to recognize the current po~,;ition and to estimate the next position of the endeffector. This tracking system can be very useful in controlling a robot at a remote site.
A multilayer feedforward neural network is employed to recognize the endeffector coveting the situation of translation, rotating and scaling types of motion. The features used to recognize the endeffector are 2D edge information from preprocessed images. The output of the neural network recognizer represents the probability of the endeffector for a specific position. The trained neural network recognizer can search for a maximum value to find the position with the highest likelihood within a limited search space. To predict the next position of the endeffector, information from the last prediction and the current position are used. Instead of analyzing data sets and modeling a prediction system, a neural network can learn the typical dynamics of the robot by way of training with patterns obtained from a series of experiments. The neural network predictor uses a smearing function to represent a real value precisely.
Combining the two ne, ural networks, for recognizing the robot endeffector and for estimating the motion, with the preprocessing stage, the whole system keeps tracking of the robot endeffector effectively with a high precision.