一种多指灵巧手内物体的位姿估计方法、装置、电子设备及存储介质
By encoding and fusing multimodal features, a global feature vector is generated. A pose regression network model is then used to estimate the pose of objects in a multi-finger dexterous hand, which solves the problem of inaccurate estimation in existing technologies and achieves higher accuracy and robustness.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- ELECTRIC POWER RES INST OF GUANGDONG POWER GRID CO LTD
- Filing Date
- 2026-06-04
- Publication Date
- 2026-07-17
AI Technical Summary
Existing object pose estimation methods cannot guarantee the accuracy of estimation in the complex wrapping grasping state of multi-finger dexterity hands. This is because they rely too much on external vision systems or single local perception modalities, resulting in fragmented perception information and an inability to establish complete spatial feature associations.
By acquiring the 3D point cloud model of the target object, the fingertip tactile images during multi-finger dexterity grasping, tactile point cloud data, palm tactile array data, and joint angle information, multimodal feature encoding and fusion are performed to generate a global feature vector, and pose estimation is performed using a pose regression network model.
It significantly improves the accuracy of object pose estimation in multi-finger dexterous hands, solves the estimation limitations caused by incomplete local perception information, and enhances prediction robustness in complex contact states.
Smart Images
Figure CN122401423A_ABST