Experimental verification of an orientation estimation technique for autonomous robotic platforms

Official URL: http://192.168.1.20/record=b1301346 (Table of Contents)

Research on autonomous platforms ranging from unmanned aerial and underwater vehicles to wheeled, tracked and legged machines enlarges the application boundaries of robotic systems. The control of these platforms, however, is a challenging task which requires the availability or estimation of various feedback variables. Orientation information of the autonomous platform is vital for robotic control. A variety of approaches and instruments are reported in the literature for orientation estimation with inertial sensors. Many approaches are based on Kalman filters. The use of Extended Kalman filters (EKF), Unscented Kalman filters and complimentary Kalman filters are proposed too. This thesis concentrates on the estimation of orientation from measurements provided by inertial sensors. The use of three-axial linear accelerometers and rate gyros is considered. The sequential use of two estimators is proposed for orientation estimation. The first one is a Kalman Filter and it is employed for the gravity estimation mainly based on acceleration readings. The second estimator has the structure of an Extended Kalman Filter which uses the gravity estimate generated by the first estimator and rate gyro readings for the orientation estimation. Orientation is estimated in a multivariable fashion, without the simplifying assumption of decoupling between onedimensional rotations about the three gyroscope axes. Therefore quite large angles of rotation can be handled accurately. The presented approach uses a quaternion representation which avoids representation singularities common with orientation descriptors like Euler angles. The computational efficiency is improved by the quaternion representation too. In order to test the estimation methods, experimental studies are carried out. A three-degrees-of-freedom robot is designed and built. The accelerometer and gyroscope unit is mounted at the tool end of this manipulator which generated test motion in the experiments. In order to create a basis for comparison, robot joint encoder data is used and actual rotation matrices during the motion are computed. The experiments indicate that the proposed technique delivers reliable orientation estimates in a large range of rotation angles and motion frequencies.