A KINEMATIC VIRTUAL POTENTIALS TRAJECTORY PLANNER FOR AUV-S Matko Barišić, Zoran Vukić, Nikola Mišković University of Zagreb, Faculty of Electrical Engineering and Computing, Laboratory for Underwater Systems and Technologies Abstract: This paper deals with trajectory planning for an autonomous, non- communicating submerged vehicle (AUV). Most maneuvering, esp. that related to obstacle avoidance, in a typical mission scenario for an AUV consists of motion at single submerged depth. Therefore a trajectory planning scheme operating in 2D has sufficient merit and applicability. A scheme for trajectory planning with cross-layer features, such as implicit inclusion of obstacle-avoidance and forming up with other moving agents, is developed in a simulated environment, at a kinematic level. The trajectory planner is based on virtual potentials, an approach that guarantees good extendibility, scalability and performance in a hard-real-time hardware-in-the-loop system. Keywords: Autonomous mobile robots, Trajectory planning, Gradients, Robot kinematics, Robot navigation 1. INTRODUCTION Trajectory planning for autonomous underwater vehicles (AUV) is a daunting challenge. Difficulties and constraints are posed by both the engineering typical of AUVs, and by the features of the environment. This problem is magnified when, rather than a single AUV, the control problem is extended to a group of AUVs. A key feature of a submerged theater of operations of such groups of AUVs is impossibility of reliable, high-bandwidth communi- cation. This precludes any shortcuts, simplifications or any method relying on at least partial communi- cation either between AUVs, or between an AUV and some form of a supervisory command, control and communication center. The autonomy of a trajectory planning method must be complete, strict and unequivocal. Also, measurement for purposes of trajectory planning in AUVs relies on high-processor- commitment operations: nonlinear filtering, applying transforms, regression and classification, on signals arriving from slow-refresh-rate sensors. Therefore the trajectory planner algorithm must take into account and manage (i.e. by multithreading and multitasking) the synchronicity between low-processor- commitment trajectory calculations and high- processor-commitment feature extraction. Section 2 explains how the stated conditions, constraints, problems and features have influenced our choice of the trajectory planning method and elaborates on the virtual potential approach. Section 3 presents the simulation results for a virtual potential method trajectory planner with problem space constrained to 2D, wherein stability problems are resolved and local minimum avoidance, obstacle- and collision-avoidance, and a limited amount of formation behavior are achieved in terms of the certain settings and modifications of the virtual potential method described in Section 2. The constraint of the problem space to 2D doesn’t theoretically flaw the arguments made, and is in turn consistent with predominant modes of usage and mission profiles of actual AUVs, where the craft are given navigation tasks at either a constant depth, or with depth controlled by a separate control loop altogether. Section 4 gives closing comments, plans for further research and surmises our findings.