Abstract
Robot manipulator Joints with only a single state sensor (either position or velocity) are limited in their ability in determining both state variables (position and velocity). Also, if the single sensor fails, the robot becomes uncontrollable. In the present paper, we propose a method which efficiently uses multi-sensor information to obtain a more precise description of joint position and velocity and allows for robust control system free from the affects of spurious sensor observation and impervious to single sensor failure. We employ an encoder and a tachometer for each joint as sensors. A technique is described for combining both uncertain sensor observations with an observation model to provide an optimal estimation of both state variables. The estimation method does not result in more than one sample period of time delay as the estimator does not use a time series analysis. In order to detect a sensor failure, we apply the cross-checking hypothesis test for redundant sensor information. In addition we can detect a control failure by including the command information of the control in the hypothesis test. The algorithm for implementing the estimation and the hypothesis test is presented. Simulation results demonstrate that the techniques works well.
