Learn to swim: Online motion control of an underactuated robotic eel based on deep reinforcement learning
As a typical underwater robot, the robotic fish has been a hot topic of research in terms of the swimming control method and the reduction of energy consumption. Body flexibility and passive designs have been proven to be effective approaches for improving the swimming performance of robotic fishes....
| Main Authors: | , , |
|---|---|
| Format: | Article |
| Language: | English |
| Published: |
Elsevier
2022-12-01
|
| Series: | Biomimetic Intelligence and Robotics |
| Subjects: | |
| Online Access: | http://www.sciencedirect.com/science/article/pii/S2667379722000262 |
| _version_ | 1811266995084066816 |
|---|---|
| author | Qixin Wang Zicun Hong Yong Zhong |
| author_facet | Qixin Wang Zicun Hong Yong Zhong |
| author_sort | Qixin Wang |
| collection | DOAJ |
| description | As a typical underwater robot, the robotic fish has been a hot topic of research in terms of the swimming control method and the reduction of energy consumption. Body flexibility and passive designs have been proven to be effective approaches for improving the swimming performance of robotic fishes. However, the addition of passive structures or motions of these approaches makes it more difficult to control the movement of robotic fish. In this paper, we proposed a deep reinforcement learning-based method for online learning control of a robotic eel with multiple passive structures. First, we designed a robotic eel with two wire-driven segments and two complaint bodies made of elastic material. Second, a simulation model of the robotic eel was built and the validity of the model was tested. Following that, the neural network was trained in simulation and deployed directly on the robotic eel, which was controlled without an underlying control model or strategy and directly online by the neural network. Finally, extensive experiments verified the effectiveness of our control method, which offers a valuable and potential solution for robots difficult to model and control. |
| first_indexed | 2024-04-12T20:54:29Z |
| format | Article |
| id | doaj.art-36ce98a8240043b08868ebc8f74c9afb |
| institution | Directory Open Access Journal |
| issn | 2667-3797 |
| language | English |
| last_indexed | 2024-04-12T20:54:29Z |
| publishDate | 2022-12-01 |
| publisher | Elsevier |
| record_format | Article |
| series | Biomimetic Intelligence and Robotics |
| spelling | doaj.art-36ce98a8240043b08868ebc8f74c9afb2022-12-22T03:17:02ZengElsevierBiomimetic Intelligence and Robotics2667-37972022-12-0124100066Learn to swim: Online motion control of an underactuated robotic eel based on deep reinforcement learningQixin Wang0Zicun Hong1Yong Zhong2Shien-ming Wu School of Intelligent Engineering, South China University of Technology, Guangzhou 510640, ChinaShien-ming Wu School of Intelligent Engineering, South China University of Technology, Guangzhou 510640, ChinaCorresponding author.; Shien-ming Wu School of Intelligent Engineering, South China University of Technology, Guangzhou 510640, ChinaAs a typical underwater robot, the robotic fish has been a hot topic of research in terms of the swimming control method and the reduction of energy consumption. Body flexibility and passive designs have been proven to be effective approaches for improving the swimming performance of robotic fishes. However, the addition of passive structures or motions of these approaches makes it more difficult to control the movement of robotic fish. In this paper, we proposed a deep reinforcement learning-based method for online learning control of a robotic eel with multiple passive structures. First, we designed a robotic eel with two wire-driven segments and two complaint bodies made of elastic material. Second, a simulation model of the robotic eel was built and the validity of the model was tested. Following that, the neural network was trained in simulation and deployed directly on the robotic eel, which was controlled without an underlying control model or strategy and directly online by the neural network. Finally, extensive experiments verified the effectiveness of our control method, which offers a valuable and potential solution for robots difficult to model and control.http://www.sciencedirect.com/science/article/pii/S2667379722000262Robotic eelDeep reinforcement learningOnline control |
| spellingShingle | Qixin Wang Zicun Hong Yong Zhong Learn to swim: Online motion control of an underactuated robotic eel based on deep reinforcement learning Biomimetic Intelligence and Robotics Robotic eel Deep reinforcement learning Online control |
| title | Learn to swim: Online motion control of an underactuated robotic eel based on deep reinforcement learning |
| title_full | Learn to swim: Online motion control of an underactuated robotic eel based on deep reinforcement learning |
| title_fullStr | Learn to swim: Online motion control of an underactuated robotic eel based on deep reinforcement learning |
| title_full_unstemmed | Learn to swim: Online motion control of an underactuated robotic eel based on deep reinforcement learning |
| title_short | Learn to swim: Online motion control of an underactuated robotic eel based on deep reinforcement learning |
| title_sort | learn to swim online motion control of an underactuated robotic eel based on deep reinforcement learning |
| topic | Robotic eel Deep reinforcement learning Online control |
| url | http://www.sciencedirect.com/science/article/pii/S2667379722000262 |
| work_keys_str_mv | AT qixinwang learntoswimonlinemotioncontrolofanunderactuatedroboticeelbasedondeepreinforcementlearning AT zicunhong learntoswimonlinemotioncontrolofanunderactuatedroboticeelbasedondeepreinforcementlearning AT yongzhong learntoswimonlinemotioncontrolofanunderactuatedroboticeelbasedondeepreinforcementlearning |