Sensor Fusion for Robotic Workspace State Estimation

Abstract

We consider the problem of tool position and orientation state estimation for robot manipulators in workspace by sensor fusion of the internal robot joint measurements with inertial measurement unit data. A prerequisite for this to be successful is accurate calibration of the sensors used. Therefore, we discuss a method for calibration of the sensor with respect to the robot end-effector, which is straightforward to apply on an arbitrary industrial manipulator. We also consider two different workspace state-estimation algorithms requiring a minimum of robot modeling; the first is based on the extended Kalman filter and the second is based on the Rao-Blackwellized particle filter. The calibration procedure and the state-estimation algorithms were evaluated and compared in extensive experiments. Both state-estimation algorithms exhibited an accuracy improvement compared to estimates provided by the forward kinematics of the robot. Moreover, both algorithms were shown to satisfy the constraints of real-time execution at 4-ms sampling period. To further evaluate and compare the robustness of the methods, the algorithms were investigated with respect to the sensitivity of the filter parameters and the noise modeling.