A Learning from Demonstration Method for Robotic Assembly with a Dual-Sub-6-DoF Parallel Robot

Abstract

Motivated by the difficulty of programming the motion of dual-arm parallel robots, an asymmetric dual-arm robot learning from demonstration (LfD) method is proposed for robotic assembly applications. Demonstration data are acquired in an indirect way with the motion capture (MoCap) system. By exploiting the stochastic formulation of Gaussian mixture model, an assembly policy is learned that models the assembly skill of specific products. Besides the LfD method with the indirect demonstration approach and the dual-arm robot of sub-6 degrees of freedom, the other contribution of this work is a dual-arm motion assignment strategy used to assign the assembly motion trajectories generated by the assembly policy to each robot arm. Redundancy of the dual-arm is utilized to deal with the problem of limited workspace. A mouse shell assembly experiment is conducted to demonstrate the usage and verify the usability of the proposed LfD method and motion assignment strategy.