Visual grasp determination and tracking in 2D dynamic scenarios

Vision-Based Grasp Tracking for Planar Objects

“…During the approaching movement of the robotic arm toward the object to be grasped, a sequence of contours is acquired in real time by the vision system. At the beginning of this sequence, as no grasp is available, an initial stable grasp is computed in the first contours (grasp synthesis) [10], [12], [13]. Then, for each new incoming contour, the algorithm tries to apply a previously computed grasp to the current image.…”

“…This involves a transfer of that grasp configuration to the current image and an evaluation of the stability of the transferred grasp, i.e., a grasp analysis [14]. The procedures for the grasp synthesis and analysis have already been presented elsewhere [10], [12] and are out of the scope of this paper. This correspondence paper focuses on three different options for the grasp transfer strategy, described in Section IV.…”

“…Thus, the tracking of the grasp points has to be performed within the tracking of the entire object. It follows that, since the object tracking method [10] has been developed for images that are correlated by translations and rotations along the normal direction to the image plane, the movements of the robotic arm are bound to these 4 DoFs.…”

“…Following the specifications given in Section III-A, this strategy is applied together with an object-tracking algorithm, described in [10], based on the use of B-splines as active contours [17]. In particular, in the contour tracking, the points sampled from the B-spline provide the RRP coordinates, i.e., the reference and interpolated points.…”

“…As an alternative to geometric approaches [7], some authors provide grasp analysis and synthesis procedures in pixel space, considering features such as the object contour [8] and using techniques such a quadtree image decomposition [9]. Nevertheless, little attention has been paid to the integration of the grasp synthesis within a visual servoing control loop [10].…”