State Estimation by Joint Approach With Dynamic Modeling and Observer for Soft Actuator

Abstract

In order to achieve a significant reduction in state estimation error and improved convergence speed, ensuring real-time responsiveness and computational efficiency, this article proposes a joint approach that combines dynamic modeling and observers to achieve accurate nonlinear state estimation of the functional soft actuator. First, inspired by the viscoelastic model, a general framework for modeling the 2D dynamics of the pneumatic network soft actuator under external conditions was studied. The dimensionless dynamic model of the soft actuator's bending deformation is derived through the dimensional analysis method. Then, an adaptive extended Kalman particle filter (aEKPF) is used for state estimation. It can restrain noise from pressure sensors and reduce drift error from rate gyroscopes. The closed-loop performance of the nonlinear pose estimation combined with the conventional control method was experimentally assessed using soft actuators and soft crawling. Results show that the aEKPF can accurately estimate the state from noise sensor measurements. Compared with conventional EKF, aEKPF improves the performance by more than 50% in terms of state estimation error and convergence speed. At the same time, in the rectilinear crawling test, the mean centroid offset in different environments is less than 3% of the soft crawling module width, verifying the effectiveness and robustness of this strategy in accurate state estimation and stability control.