Title:
NEURAL OSCILLATOR-BASED METHOD AND SYSTEM FOR CONTROLLING RHYTHMIC MOTION OF ROBOT
Document Type and Number:
WIPO Patent Application WO/2023/184933
Kind Code:
A1
Abstract:
A neural oscillator-based method and system for controlling rhythmic motion of a robot, said method comprising: acquiring the current status of a robot and the phase and frequency generated by a neural oscillator; and, according to the acquired current status, phase and frequency, and a preset reinforcement learning network, obtaining a control instruction, and controlling the robot. The preset reinforcement learning network comprises an action space, a mode-forming network, and a neural oscillator; a control structure composed of the neural oscillator and the mode-forming network can ensure the formation of a desired rhythmic motion behavior; meanwhile, the designed action space having a joint position increment can effectively accelerate rhythmic motion reinforcement learning and training processes. Thus, the problem that reward function design is time consuming and difficult in existing model-free reinforcement learning is solved.
Inventors:
ZHANG WEI (CN)
CHEN YANYUN (CN)
SHENG JIAPENG (CN)
FANG XING (CN)
TAN WENHAO (CN)
SONG RAN (CN)
LI XIAOLEI (CN)
CHENG JIYU (CN)
CHEN YANYUN (CN)
SHENG JIAPENG (CN)
FANG XING (CN)
TAN WENHAO (CN)
SONG RAN (CN)
LI XIAOLEI (CN)
CHENG JIYU (CN)
Application Number:
PCT/CN2022/125984
Publication Date:
October 05, 2023
Filing Date:
October 18, 2022
Export Citation:
Assignee:
UNIV SHANDONG (CN)
International Classes:
G05D1/08; A61B5/11
Foreign References:
CN114740875A | 2022-07-12 | |||
CN108772836A | 2018-11-09 | |||
CN110286592A | 2019-09-27 | |||
US20130138020A1 | 2013-05-30 | |||
JP2006289602A | 2006-10-26 | |||
CN113616193A | 2021-11-09 | |||
CN109866904A | 2019-06-11 |
Attorney, Agent or Firm:
JINAN SHENGDA INTELLECTUAL PROPERTY AGENCY CO., LTD. (CN)
Download PDF: