Stochastic parameters identification and localization of mobile robots

In this paper, a stochastic estimation algorithm based on a hybrid Genetic-Hidden Markov Models (GHMMs) technique is presented, with an application to nonlinear dynamic parameters identification and localization of a wheeled mobile robot. The stochastic kinematic and dynamic models of the robot and environment are introduced in order to take into account inherent uncertainties of the robot's dynamics and sensory measurements. The identification algorithm is then developed for the resulting nonlinear doubly stochastic model in a framework based on Hidden Markov Models technique. The robot state is estimated using a genetic optimization of the maximum likelihood solution. Implementation issues related to GHMMs are provided along with simulation results. Comparisons are performed and discussed with the Extended Kalman Filter for the parameters identification and state estimation problem.