Title:
ROBOTIC GRASPING PREDICTION METHOD BASED ON TRIPLET CONTRASTIVE NETWORK
Document Type and Number:
WIPO Patent Application WO/2024/087331
Kind Code:
A1
Abstract:
Disclosed in the present invention is a robotic grasping prediction method based on a triplet contrastive network. The method comprises: constructing a training set, in which each sample reflects the corresponding relationship between classification labels and tactile data at a plurality of moments during the process of a robotic arm grasping an object; training encoders on the basis of a set loss function and, during the process of training, extracting at random from the training set two samples having the same label and taking same as an anchor sample and a positive sample, and extracting at random a sample having an opposite label and taking same as a negative sample; inputting into the corresponding encoders the sample data to be encoded so as to obtain the high-dimensional feature of the anchor sample, the high-dimensional feature of the positive sample and the high-dimensional feature of the negative sample; freezing optimized parameters of the encoders, using the trained encoders to encode the input sample data into high-dimensional features, and then inputting the high-dimensional features into a classifier for training; and using the trained encoders and classifier to predict a grasping result with regard to the real-time tactile data. The present invention improves the accuracy of robotic grasping prediction.
Inventors:
YI ZHENGKUN (CN)
LIU CHENGLIANG (CN)
WU XINYU (CN)
CUI YUNDUAN (CN)
XIE XIANG (CN)
TIAN QIONG (CN)
LIU CHENGLIANG (CN)
WU XINYU (CN)
CUI YUNDUAN (CN)
XIE XIANG (CN)
TIAN QIONG (CN)
Application Number:
PCT/CN2022/137001
Publication Date:
May 02, 2024
Filing Date:
December 06, 2022
Export Citation:
Assignee:
SHENZHEN INST ADV TECH (CN)
International Classes:
B25J15/08; B25J9/16; B25J13/08; G06N3/04; G06V10/764; G06V10/80; G06V10/82
Foreign References:
CN110691676A | 2020-01-14 | |||
CN105956351A | 2016-09-21 | |||
CN112733965A | 2021-04-30 | |||
CN105005787A | 2015-10-28 | |||
CN113610151A | 2021-11-05 | |||
CN112668607A | 2021-04-16 | |||
US20190196436A1 | 2019-06-27 | |||
CN111459278A | 2020-07-28 | |||
CN113327295A | 2021-08-31 | |||
CN113168567A | 2021-07-23 | |||
CN115034286A | 2022-09-09 |
Attorney, Agent or Firm:
BEIJING ZHONG XUN TONG DA INTELLECTUAL PROPERTY AGENCY CO., LTD. (CN)
Download PDF: