Dimensionality Reduction for Motion Planning of Dual-arm Robots
Pengfei Chen, Huan Zhao, Xin Zhao, Dongsheng Ge, Han Ding · 2018
This paper solved the motion planning problem for dual-arm robots, so that the robots can move more “human-like”. In this study, firstly an Auto-Encoder (AE) method is proposed to reduce the dimensionality of the search space in the encoding process to obtain the hidden space. Then, RRT-connect in the low-dimensional space is employed to get the planning path according to the start and goal of the given task. Finally, the dual-arm robots can perform the human-like movement after mapping the samples from the low-dimensional space to the original search space. Taking advantage of the proposed method, the low-dimensional space has non-linear representation of the original space while reducing the dimensionality. Experiments are conducted by imitating pouring water using dual-arm robots. The results show that the computational load and memory are reduced significantly and movements are more human-like compared with the existing method.