Scholarlink

<p>The long-term operation of scientific payloads on the China Space Station necessitates reliable on-orbit maintenance. On-orbit maintenance is constrained by limited space, uncertain mechanical characteristics of target components, and the lack of feedback-based operational capabilities in existing control tools. These challenges highlight the need to integrate sensors into existing manipulator systems. However, the intrinsic complexity of tactile sensor data at the manipulator tool interface poses significant challenges in realizing dexterous physical interactions. This study is dedicated to overcoming the data processing limitations of integrated tactile sensors within a three-finger underactuated manipulator tool. A physics-based model of the tactile sensor and single-finger mechanics is first formulated. Based on this model, a multiple-network approach is introduced. In this approach, back propagation (BP) neural networks are used to decouple and estimate tactile features involving multiple parameters from raw sensor data. The experimental results confirm the high fidelity of the model. Specifically, all three output channels achieve <i>R</i><sup>2</sup> values above 99.5%, with a correlation coefficient of 0.997 and loss convergence to 1.3 × 10<sup>−4</sup> within 66 epochs. The real-time feedback operational capabilities of the tool are enhanced by an embedded tactile sensor. The proposed tactile perception system is integrated with force feedback and motor control loops to enable autonomous grasping with closed-loop regulation. As a result, the manipulator tool meets the stringent performance requirements for space-based manipulation. This work offers a scalable solution for tactile sensing and adaptive control in unstructured extraterrestrial environments.</p>

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

A data analysis method for tactile sensors of manipulator tools based on a BP neural network

  • 期刊论文

  • 发表日期：

    2026年4月16日

查看全文
  • Mengfei Ren,
  • Wenzhe Chen,
  • Chongfeng Zhang,
  • Jinbao Chen,
  • Chen Wang,
  • Meng Chen,
  • Jia Ma,
  • Yankai Fan

摘要

The long-term operation of scientific payloads on the China Space Station necessitates reliable on-orbit maintenance. On-orbit maintenance is constrained by limited space, uncertain mechanical characteristics of target components, and the lack of feedback-based operational capabilities in existing control tools. These challenges highlight the need to integrate sensors into existing manipulator systems. However, the intrinsic complexity of tactile sensor data at the manipulator tool interface poses significant challenges in realizing dexterous physical interactions. This study is dedicated to overcoming the data processing limitations of integrated tactile sensors within a three-finger underactuated manipulator tool. A physics-based model of the tactile sensor and single-finger mechanics is first formulated. Based on this model, a multiple-network approach is introduced. In this approach, back propagation (BP) neural networks are used to decouple and estimate tactile features involving multiple parameters from raw sensor data. The experimental results confirm the high fidelity of the model. Specifically, all three output channels achieve R2 values above 99.5%, with a correlation coefficient of 0.997 and loss convergence to 1.3 × 10−4 within 66 epochs. The real-time feedback operational capabilities of the tool are enhanced by an embedded tactile sensor. The proposed tactile perception system is integrated with force feedback and motor control loops to enable autonomous grasping with closed-loop regulation. As a result, the manipulator tool meets the stringent performance requirements for space-based manipulation. This work offers a scalable solution for tactile sensing and adaptive control in unstructured extraterrestrial environments.