Kinematic Kalman Filter (KKF) for Robot End-Effector Sensing

Abstract

In control of industrial manipulators, the position from the motor encoder has been the only sensor measurement for axis control. In this case, it is not easy to estimate the end-effector motion accurately because of the kinematic errors of links, joint flexibility of gear mechanisms, and so on. Direct measurement of the end-effector using the vision sensor is considered as a solution but its performance is often limited by the slow sampling rate and the latency. To overcome these limitations, this paper extends the basic idea of the kinematic Kalman filter (KKF) to general rigid body motion leading to the formulation of the multidimensional kinematic kalman filter (MD-KKF). By combining the measurements from the vision sensor, the accelerometers and the gyroscopes, the MD-KKF can recover the intersample values and compensate for the measurement delay of the vision sensor providing the state information of the end-effector fast and accurately. The performance of the MD-KKF is verified experimentally using a planar two-link robot. The MD-KKF will be useful for widespread applications such as the high speed visual servo and the high-performance trajectory learning for robot manipulators, as well as the control strategies which require accurate velocity information.