Title: Automatic regrouping of trajectories based on classification and regression tree

Authors: Ying Zhang; Chenguang Yang

Addresses: College of Automation Science and Engineering, South China University of Technology, Guangzhou, 510640, China ' Bristol Robotics Laboratory, University of the West of England, Bristol, BS16 1QY, UK

Abstract: Decomposing complex tasks into simple sub-trajectories can greatly reduce the difficulty of modelling and generalisation. Using dynamic movement primitive (DMP) to generalise these sub-trajectories and combining the generalised sub-trajectories in a different order can generalise the original task to a new task, which greatly improves the generalisation ability of DMP. In previous work, we manually determined the recombination order of trajectories, but this method was inefficient and time-consuming. Here, we automate the procedure with the decision tree approach. First, we use some known decision results as prior information to generate decision trees. Then, we input the starting and ending coordinates of each sub-trajectory of the new task into the decision tree. The decision tree will make decisions based on the coordinate information and choose which kind of trajectory to generalise to realise the new task. Simulation results are used to verify the effectiveness of the proposed method.

Keywords: DMP; dynamic movement primitive; trajectory segmentation; classification and regression tree.

DOI: 10.1504/IJMIC.2020.114200

International Journal of Modelling, Identification and Control, 2020 Vol.35 No.3, pp.217 - 225

Received: 03 Apr 2020
Accepted: 21 Jun 2020

Published online: 08 Apr 2021 *

Full-text access for editors Access for subscribers Purchase this article Comment on this article