Soccer robot identification using Kernel based weighted least squares

In this paper we present an application of weighted least squares WLS for identification of the dynamics of soccer robot (MIROSOT category). We used a linear model represented by state space equations to represent the dynamics of the robot and used an off-line WLS algorithm to find the coefficients of the linear model. Our main contribution is to use a kernel based (gram matrix) weight matrix to avoid the effect of the outliers in the noise acquired during the data acquisition phase for the training. The results show that this approach is better them using non-weighted algorithms.