Trajectory Correction for an AUV-mounted Underwater Robotic Manipulator Catret Mascarell J. V., Grosset D., Papadopoulos C., Andritsos F. Abstract— Newest AUVs have been proposed for inspection and maintenance tasks. These tasks imply a certain degree of autonomy also in target approaching and manipulation. This paper deals with this later issue: the manipulation in the frame of a maintenance task performed from an AUV platform. Due to high delay and low bandwidth communication systems, the robotic arm cannot be tele-manipulated from surface. The manipulator must perform autonomously, receiving from the surface control unit only high-level commands. A new reliable technique for manipulator trajectory correction, based on kinematics control, for avoiding residual motion due to a quasi-rigid docking system is presented. A priori knowledge of the target lets to use well-known computer vision techniques for target detection. Relative vehicle position and orientation can be extracted from the target geometry. I. INTRODUCTION AUVs are currently been used for exploration missions, for sampling and data collection, involving mainly autonomous navigation. Tasks as manipulation or maintenance are achieved exclusively by ROVs. This kind of robots has big constraints due to its physically linked nature. The continuous growing of ocean resources exploitations results in a bigger demand of underwater robots with less constrained capabilities. In this frame, several research european institutions are doing efforts in the way to develop a novel vehicle, included in a new underwater robot category. An AUV with full manipulation capability offers a new horizon of applications, but it offers too a wide range of problems to be solved. Communications with big time delays, that prevent a real telemanipulation, autodocking capability, with highly constrained supervision function, and autonomous manipulation ability, in an arduous environment. This paper is centered on the latest point. Once the vehicle is mechanically docked to the intervention panel, the robotic arm must perform autonomously, receiving from the surface control unit only high-level commands (i.e. TargetSelection, GoParkingPosition). In a known and static environment this can be achieved, within the manipulator operating tolerances, in simple robotic playback mode. Nevertheless, such an environment is quite difficult to achieve; it requires very rigid docking systems and purpose-built tooling and manipulator systems. Depending on important factors as current speed or vehicle hydrodynamic characteristics, this solution could be too self-assured. An alternative solution is to assume a partially rigid system and compensate for any eventual deviation from the nominal situation through a trajectory correction system. Authors are currently at the Institute for the Protection and the Security of the Citizen (IPSC) of the European Commission Joint Research Centre (JRC). Ispra (Italy). In this work, we propose a reliable trajectory correction system based on kinematic control. This technique compensates residual movements due to a quasi-rigid docking system. The use of a priori knowledge about the target, and the constrained movements of the docked vehicle, let us to develop an inexpensive and reliable trajectory correction technique. II. VEHICLE DISPLACEMENT CALCULATION The main problem for developing the trajectory correction technique resides on determining the relation between the vehicle and the panel, when this relation is not fixed. According to [1], relation between arm and target can be expressed as: () 0 6 0 6 P TCP P dev T T C t P = (1) Where 0 6 T is the homogeneous transformation matrix that describes position and orientation of the manipulator’s flange respect to the robot’s base frame. 6 TCP T is the homogeneous transformation matrix that describes position and orientation of the Tool Control Point (TCP) respect to the robot’s flange frame. ( ) 0 P C t is the homogeneous transformation matrix, that describes the panel frame respect to the robot base. P dev P is the homogeneous transformation matrix that describes the desired position and orientation, for right device manipulation, respect to the panel coordinates system. In our case, ( ) P C t 0 can be expressed as () () 1 0 0 B B P C t T C t −   =   P (2)