Abstract: The existing contact sensors for soft actuator angle measurement mainly include inertial sensors and curvature sensors. However, the measurement accuracy of inertial sensors is easily affected by the expansion of the airway embedded in the soft actuators, and the measurement of curvature sensors suffers from hysteresis and drift. To improve the accuracy of contact sensors for soft actuator angle measurement, an algorithm based on fuzzy inference and Kalman filtering is proposed for data fusion of inertial sensors and curvature sensors. Based on the BP neural network and the Long Short-Term Memory(LSTM) network, the data of curvature sensors and inertial sensors are fused respectively, which reduces the influence of the hysteresis and airway expansion of the contact sensors on soft actuators angle measurement. The experimental results show that the root mean square error accuracy of the data fusion results based on LSTM network, BP neural network, fuzzy inference and Kalman filtering are 0.51°, 0.63° and 1.59° respectively, demonstrating that the LSTM network can better improve the accuracy of the soft actuator angle measurement of the contact sensors.

Key words: soft actuator, inertial sensor, curvature sensor, Long Short-Term Memory(LSTM) network, Kalman filtering

摘要： 现有用于软体驱动器角度测量的接触式传感器主要包括惯性传感器与曲率传感器，但惯性传感器的测量精度易受软体驱动器内嵌气道膨胀的影响，曲率传感器测量则存在迟滞和漂移等问题。为进一步提高接触式传感器测量软体驱动器角度的准确性，结合模糊推理与卡尔曼滤波结合的算法实现惯性传感器和曲率传感器数据融合。基于BP神经网络和长短时记忆网络分别融合曲率传感器和惯性传感器，减少接触式传感器测量软体驱动器角度时迟滞和气道膨胀的影响。实验结果显示，采用长短时记忆网络、BP神经网络和模糊推理与卡尔曼滤波相结合的数据融合结果均方根误差精度分别为0.51°、0.63°和1.59°，表明长短时记忆网络能够更好地提高接触式传感器对软体驱动器角度的测量精度。

关键词: 软体驱动器, 惯性传感器, 曲率传感器, 长短时记忆网络, 卡尔曼滤波

CLC Number: 

• TP242