Home // ICONS 2021, The Sixteenth International Conference on Systems // View article

Comparing Kinematics-Based and Learning-Based Approaches to Robotic Arm Tasking – Using Pouring as an Example

Authors:
Tzu-Chieh Chen
Chung-Ta King

Keywords: machine learning; neural network; Inverse Kinematics; robot tasking; trajectory planning

Abstract:
The vast advances of machine learning in recent years have encouraged researchers to try learning-based end-to-end neural models for performing robotics operations. On the other hand, traditional approaches that leverage known knowledge as rules also have their merits. In this paper, we focus on robotic arm tasking and compare the learning-based end-to-end approach with a kinematics-based approach in terms of their capabilities in trajectory planning, using pouring as an example. In kinematics-based approach, object detection is obtained from a deep neural network, and arm trajectory is calculated with traditional Inverse Kinematics (IK). In the learning-based end-to-end approach, a single neural network is developed that takes RGBD images as input and outputs joint parameters of the robot arm to move the arm forward. We compare these two approaches with two scenarios, static and dynamic, in terms of their time usage and memory usage. Our experimental results show that the kinematics-based approach is more suitable for static scenarios as it uses less processing time and memory, while the learning-based approach is more suitable for complicated and dynamic scenarios.

Pages: 39 to 44

Copyright: Copyright (c) IARIA, 2021

Publication date: April 18, 2021

Published in: conference

ISSN: 2308-4243

ISBN: 978-1-61208-838-9

Location: Porto, Portugal

Dates: from April 18, 2021 to April 22, 2021