Vision-based cable separation and evaluation

The robot manipulator inserts the connector head into the designated mold and is routed through the support guides. The next phase is to perform the cable separation, where the stationary 2D camera captures the scene where the connector head is inserted, and the cable is routed. The algorithm identifies each cable distribution from their known positions in the connector and determines the point of maximum distance between the two cables of interest (between the red and white cable for this particular scenario). This position is updated to the robot manipulator, which then proceeds to the specified location, closes the fingers of the gripper and raises the arm up, performing the cable separation action. The previous steps conclude the vision-based perception and cable separation aspect of the wiring harness manipulation. After the cables have been identified and separated, the vision algorithm identifies the cable distribution again, to determine if the right cables have been separated.

Tactile-based cable routing

Tactile sensors are exploited in the cable routing task to correctly lock the connector into the specific case and to put the wires in tension during the manipulation, useful to avoid entanglement and to insert the branches into the support clips.

Wiring harness manipulation strategy

The entanglement of wire harnesses can appear at any random configuration due to the unpredictable way of putting the wire harnesses into boxes and their movement during transportation. Moreover, the entanglement of wire harnesses can contain not only simple intersections of two branches but more complex intertwisted intersections as well. Hence, we need an unambiguous and robust way of determining which branch or section of the wire harness has to be manipulated first to start the disentanglement procedure.

The video shows some examples of the manipulation actions performed during the experimental tests carried out at the UNIBO laboratory concerning the branches disentangling task.


This video shows the experimental evaluation of the bi-manual cable manipulation algorithm in a laboratory setup at Poznan University of Technology (PUT). Two UR3 arms, equipped with our own de-signed grippers were manipulating the cable in front of the RealSense D435 camera, that was providing the visual feedback. To show the robustness of the proposed algorithm, we performed the manipulation of a few different DLOs (different lengths, weights, stiffness). Also, we tested our approach in demanding situations, including occlusions and self-intersections.


The video shows the sorting of wires on the basis of estimated diameter. The diameter estimation is made by using the SVMs based classifier. Initially the correct distribution of the wires among the different diameter area is shown. Then, an operator puts the wires in a random order and with random poses between the gripper fingers. The robot sorts the wires on the table on the basis of an on-line diameter classification. All wires are correctly sorted. The task is repeated by putting the wires between the fingers with a different random sequence. The second sorting is correct anyway. Finally a comparison of initial distribution and the two obtained sorting is reported


The video shows an application of the proximity sensor to locate a wire and to estimate its shape. First, the region where the wire is placed is scanned to generate the corresponding point cloud from the sensor measurements. After the scan, the obtained point cloud is processed segment by segment to locate the points with the maximum z-coordinate, which correspond to points of the wire. Then, the x and y coordinates of the selected points are used to obtain a third order polynomial approximating the wire shape by means of a least squares method. Finally, the shape approximation is used to compute the trajectory for the robot end effector in order to follow the real wire for all its length.


The video shows side by side the tactile signals with the data related to their post processing with the estimated wire shape and the actual wire used during the experiments. Different diameters have been used.

This video compares the same data reported in the video above, but in this case superimposed, in order to allow an immediate comparison between the estimated shape and the ground truth.


The video reports an experiment of re-grasp on the basis of tactile data. After a first grasp where the wire shape is estimated, the wire is re-grasped in order to horizontally align the wire with respect to the tactile sensor frame.


The video shows the localization of the cable by the vision sensor through the Ariadne software package. The vision sensor provides the grasp pose on the cable to the robot that then execute the grasp. The system has been tested with cables of different size and colours and with different background and light conditions


This video shows a teaching process of cable routing trajectories. Trajectory and relevant poses (grasping, release…) are stored for further execution.


This video shows the preliminary test of the grasping phase of different wires (referred to section) and the sequential introduction in tray, using the OnRobot gripper with the designed fingers