Estimation of variable stiffness in flexible robot based on information assimilation

The quick and accurate estimation of variable stiffness in Adjustable Stiffness Joints (ASJ) is very important for the close loop control of the joint. On account of the stiffness is not directly measurable, and the construction of joints is complicated, the method of establishing mathematic model is not accurate enough for control system. Besides the joint models are always nonlinear, with multi-inputs and multi-outputs, and the parameters of which are coupling. All of the above factors throw down a great challenge to the effective estimation methods. Current on-line stiffness identifications are sensitive to the system conditions, the too fast or too slow changing will give rise to a bigger estimation error, and the estimation value is always deviant at the beginning time. For the purposes of establishing an effective method for control the stiffness on-line, we introduce the Extended Kalman Filter to assimilate the two pieces information of on-line identification methods and the system mathematic model, which take advantages of the both methods. As a result, the estimation accuracy of stiffness identification at the beginning time is improved, and the method is robust to the system conditions at the same time. Simulation on the variable joint model validate the approach.