Scholarlink

Locomotion Mode Recognition (LMR) is a key topic in rehabilitation and exoskeleton robotics research, with locomotion transitions presenting a particularly challenging problem. This paper proposes a CNN–LSTM hybrid neural network model for predicting lower-limb joint angles during mode transitions, capable of producing multi-step forecasts. Experimental results demonstrate that our model achieves an average mean absolute error (MAE) of 3.79 \(^\circ \) and an average lead time of approximately 90 ms. These performance levels satisfy the requirements for real-time transition prediction. Building on this predictive capability, future work will focus on using the forecasted gait parameters to recognize and classify transitions between locomotion modes.

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 CNN–LSTM-Based Prediction Method of Lower-Limb Parameters Across Multiple Locomotion Modes

  • 会议论文

  • 发表日期：

    2026年

查看全文
  • Wenke Lu,
  • Yue Ma,
  • Haoran Zhang,
  • Yichen Lin,
  • Xinyu Wu,
  • Wujing Cao,
  • Meng Yin,
  • Jianquan Sun

摘要

Locomotion Mode Recognition (LMR) is a key topic in rehabilitation and exoskeleton robotics research, with locomotion transitions presenting a particularly challenging problem. This paper proposes a CNN–LSTM hybrid neural network model for predicting lower-limb joint angles during mode transitions, capable of producing multi-step forecasts. Experimental results demonstrate that our model achieves an average mean absolute error (MAE) of 3.79 \(^\circ \) and an average lead time of approximately 90 ms. These performance levels satisfy the requirements for real-time transition prediction. Building on this predictive capability, future work will focus on using the forecasted gait parameters to recognize and classify transitions between locomotion modes.