Autonomous manipulation in underwater scenarios is a highly complex task which is still poorly studied, but with a growing interest in the last years. Two of the main phases of this problem are the detection and characterization of the object which is going to be manipulated, and the planning of this manipulation. This paper presents, on the one hand, the results of a multi-view laser reconstruction approach used in a real system. This approach consists in attaching a laser emitter and a camera in the forearm of a robotic arm. Then, moving the arm, the scene is scanned and reconstructed. This configuration allows the user to reconstruct autonomously a scene from different points of view and with a high precision. On the other hand, a grasp planning algorithm has been described. This method makes possible to autonomously detect the object of interest from a scene with multiple objects, and to choose which is the best arm configuration in order to manipulate it.