Design and Tuning of Extended Kalman Filter for Robotic System Identification

Traditional identification approaches for robotic systems based on the inverse dynamic model and the least-squares method are the most used to identify dynamic parameters of robots. However these methods often require a well-tuned filtering or estimation of the position, velocity, acceleration and torque to avoid bias in identification results. The cutoff frequency of the low-pass filter that is usually used must be well chosen, which is not always a trivial task. In this paper, we propose to use an extended Kalman filter to reduce the noise on the measured position and to estimate the velocity and acceleration. These estimates can then be fed to the controller to further reduce the noise in the control torque. The effect of the tuning of this filter is examined and the presented approach is validated through simulations and experiments on a one degree of freedom system.