Scholarlink

<p>In extreme or hazardous environments where human access is restricted, teleoperated welding offers a practical and effective solution. However, conventional teleoperation methods are typically based on direct position or velocity mapping. These approaches often suffer from limited adaptability and a strong reliance on the operator’s expertise. To overcome these limitations, we propose an intent-driven teleoperated welding system. In this system, the operator makes control decisions based on real-time weld pool images and executes them through a haptic feedback device. The system uses an intent recognition model to infer the operator’s intent and generate control commands for the welding robot. This approach combines the flexibility of human decision-making with the precision and efficiency of robotic execution. The proposed intent recognition model is built upon a variational autoencoder (VAE) framework and integrates an LSTM-attention encoder to capture spatiotemporal features in operator trajectories, thereby enhancing intent inference performance. Trajectory tracking experiments demonstrate that the proposed model reduces the mean absolute error (MAE) by 57% compared to manual operation, when both are evaluated against the same reference intent trajectory. Further welding trials demonstrate the system’s ability to assist operators in producing high-quality welds, validating the feasibility and effectiveness of the proposed teleoperated welding system.</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

Intent recognition and process control in teleoperated welding

  • 期刊论文

  • 发表日期：

    2025年10月31日

查看全文
  • Chong Peng,
  • Yanjia Yuan,
  • Dihui Chu,
  • Qiang Gao,
  • Yongzhe Li,
  • Zhuo Wang,
  • Lixin Wang,
  • Wenqing Song,
  • Xiaoyu Wang

摘要

In extreme or hazardous environments where human access is restricted, teleoperated welding offers a practical and effective solution. However, conventional teleoperation methods are typically based on direct position or velocity mapping. These approaches often suffer from limited adaptability and a strong reliance on the operator’s expertise. To overcome these limitations, we propose an intent-driven teleoperated welding system. In this system, the operator makes control decisions based on real-time weld pool images and executes them through a haptic feedback device. The system uses an intent recognition model to infer the operator’s intent and generate control commands for the welding robot. This approach combines the flexibility of human decision-making with the precision and efficiency of robotic execution. The proposed intent recognition model is built upon a variational autoencoder (VAE) framework and integrates an LSTM-attention encoder to capture spatiotemporal features in operator trajectories, thereby enhancing intent inference performance. Trajectory tracking experiments demonstrate that the proposed model reduces the mean absolute error (MAE) by 57% compared to manual operation, when both are evaluated against the same reference intent trajectory. Further welding trials demonstrate the system’s ability to assist operators in producing high-quality welds, validating the feasibility and effectiveness of the proposed teleoperated welding system.