Scholarlink

This paper proposes a vision-based tactile perception unit tailored for force sensing in the master manipulator of surgical robotic systems. The unit is capable of perceiving three-dimensional force distributions during object interaction, providing critical tactile feedback for precise and safe surgical manipulation. Based on the principle of marker-particle tactile sensing, the system features high spatial resolution, a large sensing area, and low fabrication cost. Tactile images are obtained by capturing the displacement of embedded marker particles during contact using a camera, and are subsequently used as input to a neural network to reconstruct the surface force distribution in real time. To enable large-scale data acquisition, a simulation platform was developed to automatically generate various contact scenarios. Parametric control and scripting in ANSYS were employed to simulate different contact conditions and obtain the corresponding 3D force distribution data. The fidelity of the simulated data was validated by comparing the ANSYS results with measurements from physical force sensors. The neural network adopts an encoder-decoder architecture composed of convolutional and deconvolutional layers and is trained in TensorFlow to perform force distribution estimation and contact pattern recognition. Finally, the accuracy of contact force estimation was quantitatively evaluated. A robotic end-effector integrated with the tactile perception unit was constructed, and object grasping experiments were conducted to validate the effectiveness of the proposed system.

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Simulation-Driven Learning for Vision-Based Tactile Force Reconstruction in Surgical Master Manipulators Using Random Marker Particles

  • 会议论文

  • 发表日期：

    2026年

查看全文
  • Hui Chu,
  • Xizhe Zang,
  • Peng Wang,
  • Xu Wang

摘要

This paper proposes a vision-based tactile perception unit tailored for force sensing in the master manipulator of surgical robotic systems. The unit is capable of perceiving three-dimensional force distributions during object interaction, providing critical tactile feedback for precise and safe surgical manipulation. Based on the principle of marker-particle tactile sensing, the system features high spatial resolution, a large sensing area, and low fabrication cost. Tactile images are obtained by capturing the displacement of embedded marker particles during contact using a camera, and are subsequently used as input to a neural network to reconstruct the surface force distribution in real time. To enable large-scale data acquisition, a simulation platform was developed to automatically generate various contact scenarios. Parametric control and scripting in ANSYS were employed to simulate different contact conditions and obtain the corresponding 3D force distribution data. The fidelity of the simulated data was validated by comparing the ANSYS results with measurements from physical force sensors. The neural network adopts an encoder-decoder architecture composed of convolutional and deconvolutional layers and is trained in TensorFlow to perform force distribution estimation and contact pattern recognition. Finally, the accuracy of contact force estimation was quantitatively evaluated. A robotic end-effector integrated with the tactile perception unit was constructed, and object grasping experiments were conducted to validate the effectiveness of the proposed system.