Original Article
Split Viewer
International Journal of Fuzzy Logic and Intelligent Systems 2024; 24(3): 242-257
Published online September 25, 2024
https://doi.org/10.5391/IJFIS.2024.24.3.242
© The Korean Institute of Intelligent Systems
Motion Imitation Robot Based on Artificial Neural Network with Minimization of Restrictions on Degrees of Freedom
Jeong-Hun Kang1, Seong-Jin Park2, Ye-Won Kim1, and Bo-Yeong Kang3
1Department of Artificial Intelligence, Kyungpook National University, Daegu, Korea
2Department of Mechanical Engineering, Kyungpook National University, Daegu, Korea
3Department of Robot and Smart System Engineering, Kyungpook National University, Daegu, Korea
Correspondence to :
Bo-Yeong Kang (kby09@knu.ac.kr)
This is an Open Access article distributed under the terms of the Creative Commons Attribution Non-Commercial License (http://creativecommons.org/licenses/by-nc/3.0/) which permits unrestricted noncommercial use, distribution, and reproduction in any medium, provided the original work is properly cited.
For a robot to imitate human motions, each human joint must be mapped onto the robot. In the mapping process of the NAO robot, there is a degrees-of-freedom mismatch problem between a human arm with six degrees of freedom and a robot arm with four degrees of freedom. During the collection of information on robot joint angles from human joint angles, some information on the six degrees of freedom is absent, resulting in inaccurate or erroneous movements of the robot, requiring additional calculations. In this paper, we propose a robot technology that imitates human movements by minimizing the degrees-of-freedom constraints without missing information using an artificial neural network. To verify the proposed approach, a manually measured answer dataset and an inverse kinematics answer dataset were created for each of the 919 motion frames of the human right-arm and upper-body motions. The robot imitation performance was stable through a 10-fold verification with the manually measured and inverse kinematics answer datasets for the right-arm motion imitations of 3.245◦ and 4.24◦ and the upper-body imitations of 5.10◦ and 4.82◦. In addition, as the trends of the robot prediction motion signal graph were similar to those of the answer motion signal graph, the proposed approach demonstrated a steady imitation performance.
Keywords: Artificial neural network, Motion imitation, Artificial intelligence (AI), NAO robot
Acknowledgments
Conflict of Interest
No potential conflict of interest relevant to this article was reported.
E-mail: jhkang.knu@gmail.com
E-mail: qwerty78766@naver.com
E-mail: yewonkim.knu@gmail.com
E-mail: kby09@knu.ac.kr
PDF
Standard view
Export citation
Share Article
Original Article
International Journal of Fuzzy Logic and Intelligent Systems 2024; 24(3): 242-257
Published online September 25, 2024 https://doi.org/10.5391/IJFIS.2024.24.3.242
Copyright © The Korean Institute of Intelligent Systems.
Motion Imitation Robot Based on Artificial Neural Network with Minimization of Restrictions on Degrees of Freedom
Jeong-Hun Kang1, Seong-Jin Park2, Ye-Won Kim1, and Bo-Yeong Kang3
1Department of Artificial Intelligence, Kyungpook National University, Daegu, Korea
2Department of Mechanical Engineering, Kyungpook National University, Daegu, Korea
3Department of Robot and Smart System Engineering, Kyungpook National University, Daegu, Korea
Correspondence to:Bo-Yeong Kang (kby09@knu.ac.kr)
This is an Open Access article distributed under the terms of the Creative Commons Attribution Non-Commercial License (http://creativecommons.org/licenses/by-nc/3.0/) which permits unrestricted noncommercial use, distribution, and reproduction in any medium, provided the original work is properly cited.
Abstract
For a robot to imitate human motions, each human joint must be mapped onto the robot. In the mapping process of the NAO robot, there is a degrees-of-freedom mismatch problem between a human arm with six degrees of freedom and a robot arm with four degrees of freedom. During the collection of information on robot joint angles from human joint angles, some information on the six degrees of freedom is absent, resulting in inaccurate or erroneous movements of the robot, requiring additional calculations. In this paper, we propose a robot technology that imitates human movements by minimizing the degrees-of-freedom constraints without missing information using an artificial neural network. To verify the proposed approach, a manually measured answer dataset and an inverse kinematics answer dataset were created for each of the 919 motion frames of the human right-arm and upper-body motions. The robot imitation performance was stable through a 10-fold verification with the manually measured and inverse kinematics answer datasets for the right-arm motion imitations of 3.245◦ and 4.24◦ and the upper-body imitations of 5.10◦ and 4.82◦. In addition, as the trends of the robot prediction motion signal graph were similar to those of the answer motion signal graph, the proposed approach demonstrated a steady imitation performance.
Keywords: Artificial neural network, Motion imitation, Artificial intelligence (AI), NAO robot
Fig 1.
Download
Robot motion prediction work diagram of the proposed technology for the imitation of human motion of robots and ANN construction.
Fig 2.
Download
Result loss graph of each motion with a manually measured dataset and an inverse kinematics dataset with the ANN: (a–c) inverse kinematic loss graph (walking-jumping-greeting) and (d–f) manually measured loss graph (walking-jumping-greeting).
Fig 3.
Download
Dataset configuration.
Fig 4.
Download
Robot imitation of human right-arm motion (left, human motion; right, robot motion) and result graph of each motion with the manual measurement dataset and inverse kinematics dataset: (a–c) right-arm motion shape (walking-jumping-greeting), (d–f) right-arm signal answer with inverse kinematics, (g–i) right-arm signal prediction with inverse kinematics, (j–l) right-arm signal answer with manual measurement, and (m–o) right-arm signal prediction with manual measurement. The x-axis represents time, and the y-axis represents the angle.
Fig 5.
Download
Robot imitation of human upper-body motion (left, human motion; right, robot motion) and result graph of each motion with the manual measurement dataset and inverse kinematics dataset: (a–c) upper-body motion shape (walking-jumping-greeting), (d–f) upper-body signal answer with inverse kinematics, (g–i) upper-body signal prediction with inverse kinematics, (j–l) upper-body signal answer with manual measurement, and (m–o) upper-body signal prediction with manual measurement. The x-axis represents time, and the y-axis represents the angle.
Fig 6.
Download
Figure A1. Result graph of each motion with the manual measurement dataset and inverse kinematics dataset showing the difference between the answer and predicted angle values: (a–c) difference in the joint angles of the right arm with inverse kinematics, (d–f) difference in the joint angles of the right arm with manual measurement, (g–i) difference in the joint angles of the upper body with inverse kinematics, and (j–l) difference in the joint angles of the upper body with manual measurement. The x-axis represents time, and the y-axis represents the angle.
Fig 7.
Download
Figure A2. Robot imitation of human right-arm motion (left, human motion; right, robot motion) and result graph of each motion with the manual measurement dataset and inverse kinematics dataset, trained for 5,000 epochs: (a–c) right-armmotion shape (walking-jumping-greeting), (d–f) right-arm signal answer with inverse kinematics, (g–i) right-arm signal prediction with inverse kinematics, (j–l) right-arm signal answer with manual measurement, and (m–o) right-arm signal prediction with manual measurement. The x-axis represents time, and the y-axis represents the angle.
Fig 8.
Download
Figure A3. Robot imitation of human right-arm motion (left, human motion; right, robot motion) and result graph of each motion with the manual measurement dataset and inverse kinematics dataset, trained for 15,000 epochs: (a–c) right-arm motion shape (walking-jumping-greeting), (d–f) right-arm signal answer with inverse kinematics, (g–i) right-arm signal prediction with inverse kinematics, (j–l) right-arm signal answer with manual measurement, and (m–o) right-arm signal prediction with manual measurement. The x-axis represents time, and the y-axis represents the angle.
Fig 9.
Download
Figure A4. Robot imitation of human upper-body motion (left, human motion; right, robot motion) and result graph of each motion with the manual measurement dataset and inverse kinematics dataset, trained for 5,000 epochs: (a–c) upper-body motion shape (walking-jumping-greeting), (d–f) upper-body signal answer with inverse kinematics, (g–i) upper-body signal prediction with inverse kinematics, (j–l) upper-body signal answer with manual measurement, and (m–o) upper-body signal prediction with manual measurement. The x-axis represents time, and the y-axis represents the angle.
Fig 10.
Download
Figure A5. Robot imitation of human upper-body motion (left, human motion; right, robot motion) and result graph of each motion with the manual measurement dataset and inverse kinematics dataset, trained for 15,000 epochs: (a–c) upper-body motion shape (walking-jumping-greeting), (d–f) upper-body signal answer with inverse kinematics, (g–i) upper-body signal prediction with inverse kinematics, (j–l) upper-body signal answer with manual measurement, and (m–o) upper-body signal prediction with manual measurement. The x-axis represents time, and the y-axis represents the angle.
-
Table 1 . Average error of 10-fold cross-validation and test set validation for right-arm motion imitation (unit: degree).
10-fold test Test set Manually measured_Proposed 3.2450 ± 0.5 2.9743 ± 0.5 Inverse kinematics_Proposed 4.2357 ± 0.5 4.6399 ± 0.5
-
Table 2 . Average error of 10-fold cross-validation and test set validation for upper-body motion imitation (unit: degree).
10-fold test Test set Manually measured_Proposed 5.1018 ± 0.5 5.3845 ± 0.5 Inverse kinematics_Proposed 4.8150 ± 0.5 4.6146 ± 0.5
Article Tools
Metrics
Share this article on :
Related articles in IJFIS
-
Grey Wolf Optimization-Based Artificial Neural Network in the Development of an Automated Heart Disease Prediction Model
Amirthalakshmi Thirumalai Maadapoosi, Velan Balamurugan, V. Vedanarayanan, Sahaya Anselin Nisha, and R. Narmadha
International Journal of Fuzzy Logic and Intelligent Systems 2024; 24(3): 231-241
https://doi.org/10.5391/IJFIS.2024.24.3.231
-
RLC Fault Detection Based on Image Processing and Artificial Neural Network
Ali Rohan and Sung Ho Kim
International Journal of Fuzzy Logic and Intelligent Systems 2019; 19(2): 78-87
https://doi.org/10.5391/IJFIS.2019.19.2.78
-
Application of Computational Intelligence Techniques to an Environmental Flow Formula
Zong Woo Geem, and Jin-Hong Kim
International Journal of Fuzzy Logic and Intelligent Systems 2018; 18(4): 237-244
https://doi.org/10.5391/IJFIS.2018.18.4.237
Grey Wolf Optimization-Based Artificial Neural Network in the Development of an Automated Heart Disease Prediction Model
Amirthalakshmi Thirumalai Maadapoosi, Velan Balamurugan, V. Vedanarayanan, Sahaya Anselin Nisha, and R. Narmadha
International Journal of Fuzzy Logic and Intelligent Systems 2024; 24(3): 231-241 https://doi.org/10.5391/IJFIS.2024.24.3.231RLC Fault Detection Based on Image Processing and Artificial Neural Network
Ali Rohan and Sung Ho Kim
International Journal of Fuzzy Logic and Intelligent Systems 2019; 19(2): 78-87 https://doi.org/10.5391/IJFIS.2019.19.2.78Application of Computational Intelligence Techniques to an Environmental Flow Formula
Zong Woo Geem, and Jin-Hong Kim
International Journal of Fuzzy Logic and Intelligent Systems 2018; 18(4): 237-244 https://doi.org/10.5391/IJFIS.2018.18.4.237
