Using a Flexible Skill-Based Approach to Recognize Objects in Industrial Scenarios

Abstract

Traditional industrial robots are highly efficient and precise and therefore well suited for carrying out simple, repetitive tasks. They are, however, complicated and time consuming to setup and re-program to perform new tasks. Skill-based programming attempts to reduce both the required time as well as the need for highly specialized staff for setting up modern collaborative robots. This paper proposes a skill for recognition and classification of different objects. The skill is parameterized using manual kinesthetic teaching, and machine learning based on SIFT features, Bag of Words, and SVM is used to classify objects. A user study with 20 test participants shows that robotics novices after only a short introduction are able to instruct the skill and combine it with other skills (pick and place) to program a complete task.