Object Detection in Motion Reproduction System with Segmentation Algorithm

Human skillful motion can be extracted through teleoperation. The saved motion can be reproduced by robots without human which is known as motion reproduction. If the environment is changed after the motion is saved, the saved motion is required to be modified according to the target object. Traditionally, the target object is specified by human operator. The proposed method provide a algorithm which automatically detects the target object in the image based on the saved motion. The object detection is examined through actual saved motion data. The target object extracted from image based on a single saved motion can be used as template for other reproduction. The proposed method is also believed to reduce the burden of annotation for save motion.

I. Introduction

Several human works are being replaced by robots as technology advances. However, unstructured complex tasks such as surgery, harvesting, and cooking are still challenging for robots. The key for the robotic automation is planning motion. The planning motion includes manual programming and learning from demonstration (LfD) [1]–[3].