An improved approach of task-parameterized learning from demonstrations for cobots in dynamic manufacturing

Shirine El Zaatari, Yuqi Wang, Yudie Hu, Weidong Li

Research output: Contribution to journalArticlepeer-review

Abstract

Task-Parameterized Learning from Demonstrations (TP-LfD) is an intelligent intuitive approach to support collaborative robots (cobots) for various industrial applications. Using TP-LfD, human’s demonstrated paths can be learnt by a cobot for reproducing new paths for the cobot to move along in dynamic situations intelligently. One of the challenges to applying TP-LfD in industrial scenarios is how to identify and optimize critical task parameters of TP-LfD, i.e., frames in demonstrations. To overcome the challenge and enhance the performance of TP-LfD in complex manufacturing applications, in this paper, an improved TP-LfD approach is presented. In the approach, frames in demonstrations are autonomously chosen from a pool of generic visual features. To strengthen computational convergence, a statistical algorithm and a reinforcement learning algorithm are designed to eliminate redundant frames and irrelevant frames respectively. Meanwhile, a B-Spline cut-in algorithm is integrated in the improved TP-LfD approach to enhance the path reproducing process in dynamic manufacturing situations. Case studies were conducted to validate the improved TP-LfD approach and to showcase the advantage of the approach. Owing to the robust and generic capabilities, the improved TP-LfD approach enables teaching a cobot to behavior in a more intuitive and intelligent means to support dynamic manufacturing applications.
Original languageEnglish
Pages (from-to)(In-press)
JournalJournal of Intelligent Manufacturing
Volume(In-press)
Early online date6 Feb 2021
DOIs
Publication statusE-pub ahead of print - 6 Feb 2021

Funder

This research is funded by the Coventry University, the Unipart Powertrain Application Ltd. (U.K.), the Institute of Digital Engineering, U.K., and a research project sponsored by the National Natural Science Foundation of China (Project No. 51975444)

Keywords

  • Collaborative robots
  • Learning from demonstration
  • Reinforcement learning

ASJC Scopus subject areas

  • Software
  • Industrial and Manufacturing Engineering
  • Artificial Intelligence

Fingerprint Dive into the research topics of 'An improved approach of task-parameterized learning from demonstrations for cobots in dynamic manufacturing'. Together they form a unique fingerprint.

Cite this