Robust Visual Servo Control and Tracking for the Manipulator of a Planetary Exploration Rover

Kurzfassung

To collect samples and handle tools, planetary exploration rovers commonly employ light-weight robotic manipulators. These can suffer from undesirable positioning imprecision due to erroneous end-effector pose estimates obtained by the manipulators kinematics, leading to the failure of the manipulation task. This thesis presents a vision-based end-effector pose correction pipeline to improve the positioning precision of the end-effector during manipulation tasks. Our approach corrects the end-effector pose by fusing the estimates obtained by the manipulators kinematics with information obtained from monocular vision data. We propose a gradient based method to track a set of active markers within the image stream, which provides us with additional information on the covariance of the retrieved image points. In order to recover the 3D pose estimate of the end-effector, we make use of the maximum likelihood perspective-n-point algorithm, allowing us to propagate the image point uncertainties to their 3D pose covariances. Based on evaluations using recorded ground-truth data, we show that our tracking method leads to a reduction of the kinematic position error by up to 77%. To operate outdoors and under changing illumination conditions, the robustness of the tracking approach is paramount. Based on the propagated covariance information, we employ an Error-State Kalman filter for the rejection of pose outliers and the reduction of pose jitter. Its smoothing capabilities are confirmed in simulation. We further show the application of the vision-based correction pipeline as part of a visual servoing scheme designed for the collection of payload boxes by the manipulator of the Lightweight Rover Unit 2, developed at the Institute for Robotics and Mechatronics of the German Aerospace Center. We propose a switching control scheme that applies a position-based visual servo (PBVS) for movements of the end-effector in free space and switches to a PBVS based hybrid impedance visual servoing scheme for movements in close proximity or in direct contact with the coupling partner, to ensure the safe interaction between the manipulator and the payload. The implemented PBVS approach is lastly evaluated in simulation. We show its successful execution and stability in the vicinity of singularities as well as the avoidance of joint position and velocity limits.