Sensorless Force Estimation of a Lever-Based Variable Stiffness Actuator Using a Current-Deflection Fusion Method

Sensorless force estimation of a variable stiffness actuator (VSA) is crucial for safe physical human-robot interaction. The current-based method suffers from inaccurate model parameters and susceptibility to external disturbances, whereas the deflection-based method lacks a reliable estimation of dynamic stiffness cases. This article presents a sensorless force estimation scheme using a current-deflection fusion (CDF) method for a lever-based VSA. First, unified dynamic models of the driving systems are established, and the unknown parameters are identified before employing the current information. Taking into account the frictions of all the transmission components, the developed unified dynamic models enhance the performance of parameter identification. Then, the CDF-based force estimation is developed with an adaptive coefficient. It effectively improves the estimation accuracy in both quasi-static and dynamic stiffness situations and degrades sensitivity to external disturbances. Extensive experiments are conducted to demonstrate the effectiveness of parameter identification with the developed unified dynamic models and the estimation quality of the proposed CDF method, through evaluating the absolute error of estimated force.