Topological State Recognition And Untangling Operations For Unknown Number Of Deformable Linear Objects

The operation of deformable objects,especially deformable linear object(DLO)s,represents an open challenge in the field of robotics.Capability of recognizing the complex topological state of DLO is the key to solve the problem.In this paper,a solution based on deep learning is proposed.It is applied to the scenarios where multiple DLOs with unknown number are placed under complex background.The proposed method is used to recognize both the self-tangling of one single DLO but also the tangling between different DLOs.A strategy for untangling the multiple DLOs is proposed.The effectiveness of the proposed method was verified by using a dual-arm robot system.Its results demonstrated that even if the exact number of the DLOs are not known,the robot can successfully eliminate the tangling existing within the DLOs.In addition,with the help of Deep Learning,this approach demonstrated its robustness to unknown background,which is expected to be useful in practical applications.The endpoints out of the field of view caused by limited camera field of view is solved in this paper.The topology state recognition method and untangling operation strategy are validated using a two-armed robot platform.The performance tests of topology recognition methods and untangling operations under different conditions are designed.Experimental results show that the proposed topology state recognition method is effective and reliable,the robot can accurately recognize the topology state of DLOs and untangle them,the operations can be correctly performed automatically in the condition of DLOs in different simplist status.The program can terminate correctly without human intervention.