Topological State Recognition For Deformable Linear Objects Untangling Conducted In Unknown Background

With the increasing level of automation in industrial production assembly,there is an increasing demand for autonomous operation of flexible bodies.Deformable linear objects(DLO)is an important component of the flexible body,and there are many differences compared to the planar flexible body and the bulk flexible body.DLO has a complicated winding state,the same topological state corresponds to a plurality of different geometric states and the actual operation also has a complicated background,which lead to the manual operation of the DLO such as cable,rope,hose,etc.in the current assembly,which cannot be manually operated by the robot.Therefore,autonomous operation of DLO becomes a critical step in improving the level of automation.Based on the above background,this paper proposes a method based on deep learning to identify the topological state of deformable linear objects and a untangling strategy for topological state,and the topological state of deformable linear objects is untangled through the mechanical arm.DLO's winding is divided into a single self-wound and multiple heel winding.Therefore,for the single and double DLO,the target detection algorithm is first used to obtain the endpoints and cross points of DLO.A color gradient map of the DLO is then obtained using a convolutional neural network.For each root using a different gradient color,by analyzing the pixel values of the local area of the intersection of the color gradient map,the upper and lower relationship of the intersection segment and the order of each cross point passing through from beginning to end are obtained.Thus,the topological state of the single and double DLO is obtained.Next,for the untangling of the topological state of DLO,three basic operations that can simplify the topology state are used to untangle it into the simplest state.Specifically,by analyzing the sequence of cross points that have been obtained in the topology state recognition,a part that can perform three basic operations is obtained.With the pixel values of the partial intersections are analyzed,the segment to be operated by untangling is determined.After the mathematical calculation,the clamping point at the time of untangling is determined,and the untangling step is planned according to the three basic operation categories.So far,the untangling strategy for the topology state is obtained.Finally,the dexterous hand is used to grasp DLO and the KUKA arm moves to realize the simplest state of untangling the DLO.The proposed method has good robustness to the complex background,large variation of illumination environment and limited camera field of view in actual production assembly.Finally,the feasibility of identification and untangling strategy is verified by the mechanical arm untangling experiment.