An Online Stiffness Estimation Approach for Variable Stiffness Actuators Using Lever Mechanism

The lever mechanism is widely used in variable stiffness actuators for its advantages of a wide stiffness regulation range and fast response. However, online stiffness estimation for this mechanism has not been fully investigated. The high nonlinearity of joint flexibility and the dynamic coupling between the deflection angle and position of the stiffness regulation motor both hinder the improvement of estimation accuracy. This paper proposes an alternative online stiffness estimation method to address the above problems. First, a nonlinear observer with low latency is constructed to estimate the flexibility torque from noisy measurements. Then, by introducing the prior information around the equilibrium position, a reduced parametric model is constructed to approximate the dynamics of the flexibility torque. On this basis, the flexibility torque is identified using a recursive least square algorithm and the stiffness is calculated from its first-order derivative. The simulation and experimental results validate that the estimation accuracy of the proposed method is improved compared with the traditional residual based method, especially in poor excitation conditions such as high stiffness and light load. Furthermore, the computational cost is also reduced due to the compact structure of the parametric model.