Wiring Connector-Terminated Cables Based on Manipulation Planning with Collision-Free EMD Net

Kimitoshi Yamazaki(*), Kyoto Nozaki(*), Yuichiro Matsuura(**), Solvi Arnold(*)

(*)Shinshu University, (**)SEIKO-EPSON Corp.

Abstract: In this paper, we propose a method of manipulation planning for cable wiring required at assembly of electric appliances etc. We assume that a robot grasps a connector attached to the end of a cable and makes a motion to bring that connector closer to the socket. To accomplish it automatically, we propose a novel manipulation planning method. The method uses neural networks with extension of EMD net, which can predict shape changes of deformable object and therefore can generate robot’s motion sequence to give the deformable object with desired shape changes. This enables to find a connector moving path that avoids collision between the cable with surrounding environment. We conducted experiments under several different length of cables. There, some functions required for actual wiring, e.g., online cable shape modification, are also introduced. Experimental results show that actual cable manipulation can stable be achieved by using the proposed method

Keyword: wiring, manipulation planning, deformable linear object, EMD net

Method

  • Preparing

Dataset

We can provide the cable simulation data for neural network training as follows. Please refer to README.txt and contact us (kyamazaki at shinshu-u.ac.jp) if you interested in.

  • README.txt
  • 150 mm length (600MB)
  • 200 mm length (630MB)
  • 250 mm length (600MB)
  • 300 mm length (870MB)
  • 350 mm length (730MB)

Experiment movie