Behaviour based visual servo controller for satellite capture

on Advanced Space Technologies for Robotics and Automation 'ASTRA 2002' ESTEC, Noordwijk, The Netherlands, November 19 - 21, 2002
Figure 1 Approach envelope for SSRMS capturing a free-flyer [7] During capture of a free-flying satellite an astronaut controls the arm using images from the arm cameras and moves the end-effector through a sequence of way points defined relative to the satellite. The last waypoint before the capture is 1.5-2m away from the interface. At this distance the arm operator moves the end-effector so as to reduce the lateral offsets and orientation angles to zero. The capture is achieved with a single motion along the camera/end-effector axis, which minimises the time that the end-effector spends in the close proximity of the free-floating satellite. 2.2 Autonomous satellite capture Capturing free-floating satellites autonomously will most likely follow procedures developed for manned operations, and specifically the concepts of way points, incremental alignment and stay-out zones. A vision system will process intensity or range images and estimate pose (position and orientation) of the satellite and the interface. The cameras may be mounted on the servicer and/or the manipulator. An end-effector camera will be used for the close range operations. The initial distance between the end-effector and target spacecraft will depend on the arm size and design, and whether the spacecraft control will be integrated with the arm control, and will range from several to tens of meters. The final position corresponds to a distance between the end-effector camera and the satellite surface in the captured position (0.2-0.5m). During the operation the vision system will have to fixate the cameras on the observed object ensuring that neither the satellite drift or end-effector motion could cause loss of visibility and data. 3 VISUAL SERVOING APPROACHES Visual servoing is typically classified into position based, image based and hybrid servoing [4]. All of the approaches attempt to minimise an error between the set and measured values. However, they differ in which error function is minimised and the type of measurements performed by the vision system. A thorough analysis and comparison of these methods is presented in [6]. Image based methods operate by extracting image features, matching them with their desired locations, and minimising the image plane error between them. Pose based methods minimise an error between measured and required pose in the Cartesian space. Image based methods appear attractive as their computational cost is lower - it is not necessary to compute the 3D pose of the observed object. However, the visual servoing task must be expressed as a sequence of feature locations in images instead of a series of Cartesian frames of reference. In our work we rely on position based visual servoing approach as full pose information (6 dof) is computed by our vision systems. 4 MDR VISION SYSTEM A vision system for satellite proximity operations, which include satellite capture using a manipulator, is described in a companion paper [5]. This vision system uses stereo cameras and a satellite model to compute pose of the observed satellite if it is within stereo range (medium range). At a short range (below the stereo range) the vision system operates in a monocular mode. The vision system automatically acquires initial pose of the satellite (or interface) and tracks it in subsequent images. The system automatically reconfigures itself during the operation depending on the distance and relative orientation between the satellite and cameras. The prototype vision system tracks satellite pose at the rate of 2Hz in the medium range (1.2m - 5m) and 12 Hz in the short range (below 1.5m) [5]. The system operates successfully under expected space conditions (illumination, space materials, viewing angles and distances), that cause partial data loss due to shadows, occlusions and specular reflections. At each phase of the operation the vision system computes confidence of estimated values. Low confidence may trigger reacquisition of pose and continuation of tracking, altering the controller behaviour or aborting the operation.
Michael Liu, Piotr Jasiobedzki
MD Robotics, 9445 Airport Rd., Brampton ON, Canada, L6S 4J3 {mliu, pjasiobe}@mdrobotics.ca
1 INTRODUCTION Unmanned space servicing requires an