Electromyogram (EMG)-based continuous joint motion estimation is essential in various human-involved scenarios, and variations across speeds and subjects have been a long-time challenge. This article proposes a novel adaptive oscillator-driven gait phase model for continuous motion estimation of lower limbs and achieves robust performance in miscellaneous walking speed conditions. The proposed model consists of three key components. First, the mapping relationships with the state of the gait phase and its time derivative (gait frequency) are analyzed and constructed by the Gaussian process regression (GPR). With these relationships, the subject-specific profiles of the gait phase state and the joint motion cross speeds can be formulated. Then, an improved adaptive frequency oscillator (AFO) with EMG activation signals is designed to online estimate the inputs of the gait phase profiles, and an additional phase alignment module is elaborated to exponentially compensate for the offsets and errors. Afterward, the proposed model integrates the information derived from both the formulated position profile and estimated integral of the velocity with an adaptive unscented Kalman filter (UKF) scheme to reject disturbance, thus improving the accuracy and smoothness of the estimation outputs of joint motion. Comprehensive validation experiments were conducted to estimate the motion of three joints (hip, knee, and ankle) on both a benchmark dataset and a self-collected dataset. Compared with the existing methods, the proposed method achieved significantly better performance ( $p < 0.01$ ) in terms of root mean square error (RMSE), correlation coefficient (CC), and coefficient of determination ( $R^{2}$ ). Furthermore, the standard deviations of RMSE decreased 31.8% over 17 subjects and 36.3% over 28 speed conditions in the benchmark dataset, and decreased 38.1% and 42.7% in the self-collected dataset, both of which demonstrated good consistency in the cross-subject and cross-speed estimation of three joints. These results indicate that the proposed adaptive oscillator-driven gait phase model promises an accurate and robust solution for joint motion estimation in miscellaneous walking conditions.

中文翻译：

---

用于跨速度连续运动估计的自适应振荡器驱动步态阶段模型

基于肌电图 (EMG) 的连续关节运动估计在各种涉及人类的场景中至关重要，而速度和对象之间的变化一直是一个长期的挑战。本文提出了一种新颖的自适应振荡器驱动步态相位模型，用于下肢的连续运动估计，并在各种步行速度条件下实现了稳健的性能。所提出的模型由三个关键部分组成。首先，通过高斯过程回归（GPR）分析并构建步态相位状态与其时间导数（步态频率）的映射关系。利用这些关系，可以制定特定于受试者的步态阶段状态和关节运动交叉速度的轮廓。然后，设计了一种带有 EMG 激活信号的改进自适应频率振荡器（AFO）来在线估计步态相位轮廓的输入，并精心设计了一个附加的相位对齐模块来以指数方式补偿偏移和误差。随后，所提出的模型将来自公式化位置轮廓和速度估计积分的信息与自适应无迹卡尔曼滤波器（UKF）方案相结合以抑制干扰，从而提高关节运动估计输出的准确性和平滑度。进行了全面的验证实验，以估计基准数据集和自行收集的数据集上三个关节（髋部、膝部和踝部）的运动。与现有方法相比，所提出的方法取得了明显更好的性能（ $p < 0.01$ ）以均方根误差 (RMSE)、相关系数 (CC) 和决定系数 ( $R^{2}$ ）。此外，基准数据集中 17 个受试者的 RMSE 标准差下降了 31.8%，28 个速度条件下的 RMSE 标准差下降了 36.3%，自收集数据集中的 RMSE 标准差下降了 38.1% 和 42.7%，两者在跨受试者中表现出良好的一致性。以及三个关节的交叉速度估计。这些结果表明，所提出的自适应振荡器驱动步态相位模型为各种步行条件下的关节运动估计提供了准确且鲁棒的解决方案。