Robust Estimation of Variable Stiffness in Flexible Joints

We consider the problem of estimating on line the nonlinear stiffness of flexible transmissions in robot joints, with special reference to actuation devices with adjustable stiffness in serial configuration. These joints are characterized by a principal motor for controlling the link motion and secondary motor for adjusting the stiffness. In this actuation configuration, the flexible transmission undergoes relatively small deformations and the stiffness estimation problem is more challenging due to poor excitation conditions. We improve our previous general approach, combining a residual-based flexibility torque estimator that uses also a kinematic Kalman filter to handle discretization and quantization errors with an enhanced recursive least squares algorithm that does not suffer from lack of persistent excitation. As a result, stiffness is estimated in a more robust way using only position measurements on the motor sides and motor dynamic parameters. The performance of the proposed estimation method is illustrated through simulations and experiments on the AwAS joint developed at IIT.