Robust Tracking of Longitudinal Objects Lying on the Sea Floor with an AUV equipped of a Scanning Sonar Christian Barat Laboratoire I3S (Informatique, Signaux et Systèmes de Sophia Antipolis) 2000 route des Lucioles, BP 121, 06903 Sophia Antipolis cedex, FRANCE barat@i3s.unice.fr Maria João Rendas Laboratoire I3S (Informatique, Signaux et Systèmes de Sophia Antipolis) 2000 route des Lucioles, BP 121, 06903 Sophia Antipolis cedex, FRANCE rendas@i3s.unice.fr Abstract - The paper considers the problem of using an autonomous underwater platform equipped of a scanning profiler sonar to track elongated objects lying on the sea floor, e.g. pipelines. The algorithm presented assumes the existence of features that enable the discrimination between sonar profiles received from the tracked object and those received from the background sea bottom. The algorithm does not require explicit classification of the profiles received (as coming from the tracked object or from the sea bottom), and does not rely on reconstruction of the geometry of the surface of the tracked object (alowing tracking of proud or partially buried objects). Simulation results illustrate the performance of the algorithm presented, as well as the effect of departures from the assumed geometric and reflectivity characteristics of the actual tracking configuration. Results of processing of real sonar profiles obtained at sea with a man-madeobject of cylindrical geometry are presented, demonstrating the appropriateness of the approach proposed. I. INTRODUCTION This paper proposes a new criterion for autonomous tracking of elongated objects lying on the sea floor with a profiler sonar scanning the vertical plane orthogonal to the robot’s direction of motion. Contrary to the majority previously proposed tracking strategies [5,6] – mainly relying in the use of either video cameras, or of muti-beam or side-scan sonar sensors -- the criterion presented here does not rely on the reconstruction, from the sonar measurements, of the local relative robot/object geometry, and considers the use of a simple single-beam mechanically scanning sonar. Obviously, the approach proposed can be transposed to more sophisticated multi-beam sonar systems, with considerable improvemens in the quality and efficiency of tracking. Instead of trying to fit geometric models of the tracked object to distinctive features of the data, and infer from their location the current tracking offset, our tracking approach relies on mantaining a proper balance between the statistical characteristics of the sonar returns sensed from both halves (left and right) of the vertical plane scanned by the sonar. We show that, under mild conditions, driving to zero the tracking error proposed is equivalent to keeping the conic sector scanned by the sonar centred on the object’s axis. The computation of the error criterion is entirely based on received sonar data, requiring only an initial step to learn the characteristics of the sea bottom in areas close to the object. The paper is organised as follows. In the next section we present the platform used in the study and the characteristics of the acoustic sensor on which tracking is based. In section III we present and motivate our tracking criterion, proposing an error measure on the basis of which the reference for a low-level rate controller is generated. Section IV presents simulation results that illustrate the performance and limitations of the approach proposed. Results on real signals acquired at sea are presented in Section V, confirming the appropriateness of our guidance criterion, and its robustness with respect to assumed modeling conditions. Finally, in Section VI we summarize the contributions of the paper. II. PLATFORM AND SENSOR A Platform Fig 1: The Phantom being launched at sea. The underwater platform used in this study is the ROV Phantom, 1 shown in Figure 1. This robot is equipped of three thrusters, two allowing control in the horizontal plane (forward/reverse, turning) and another controlling 1 Phantom is a Remotely Operated Vehicle (ROV) produced by Deep Ocean Engineering, USA, which has been made available for research in underwater robotics at I3S through a special educational arrangement. Sonar