3D motions planning of humanoid arm using learned patterns

Abstract

Humanoid arm has a wide range of applications such as automatic assembly and welding. Due to its complex and nonlinear properties, it is difficult to achieve high robustness and fast response synchronously for the motion planning of humanoid arm. Very recently, it has been proved that imitating human movement system can improve the performance of robot control [11]. This paper proposes a new 3D motion planning method of humanoid arm based on habitual planning theory. The method we proposed is a pre-training algorithm to map the target inputs into a series of patterns of the 3D motion space. Therefore, our proposed method can realize 3D motion planning of humanoid arm. The simulation experimental results demonstrate that our proposed method can use a finite number of patterns (143 patterns used in our experiment) to cover most areas (more than 99%) of the 3D motion space of humanoid arm.