Discriminative parameter determination of divided difference filter for mobile robot localization

In this paper, we propose a learning method to solve the parameter determination problem of divided difference filter (DDF) for accurate localization. DDF can achieve comparatively accurate localization than other Kalman filter algorithms in poor GPS area. However, parameter determining process of DDF requires significant time and engineering cost. Furthermore, it is difficult to obtain optimal parameters for accurate localization by hand-tuning. DDF has three parameters which should be determined: covariance matrices of input and measurement noise and a Hyper-parameter. Our technique uses a discriminative learning method to determine these parameters. The proposal method absolves developers from the cumbersome process of parameter setting. This paper describes the efficiency of our technique through simulations and an experiment.