A Data-Driven Multi-Scale Online Joint Estimation of States and Parameters for Electro-Hydraulic Actuator in Legged Robot

In order to satisfy the real-time need of model-based controllers for model parameters and full states feedback, this paper has conducted in-depth research on the states and parameters estimation of electro-hydraulic actuator in legged robot with three problems for time-varying parameters estimation (including system parameters and external load force), non-measurable states estimation and measurable states filtering. The first-order trajectory sensitivity method based on the dynamic model is used to determine the parameter set to be estimated, and the parameter fast and slow characteristics are analyzed in detail to obtain the generalized states and slow-varying parameters. Then, the combined algorithm with a fast-varying time scale (composed of a fusion kalman filter and a fast-varying time scale extended kalman filter) and a slow-varying time scale (composed of a slow-varying time scale extended kalman filter) is innovatively proposed to realize the data-driven multi-scale online joint estimation of states and parameters for the actuator system. Finally, the results of three comparative experiments show that the proposed algorithm has better stability, faster convergence speed and more accurate estimation than the dual extended kalman filter algorithm, and the states and parameters estimated by the proposed algorithm accurately reflect the actual characteristics of actuator. Moreover, the algorithm has strong adaptability and robustness in different actuator hardware environment and strong convergence ability for different initial values of states and parameters.