Localization of mobile robot using discrete space particle filter

A discrete space based particle filter (DSPF) is presented for mobile robot localization based on the thought that the particle filter efficiency mainly depends on the updating of particle set. According to the localization system error, the robot running environment is divided into variable precision grids and discrete system models are described by these grid-particles. The laser scanning data of these grid-particles are acquired and pre-stored as particle characteristics at the environment map pre-processing stage. Particles are discretely approximated to fixed grid-particles and the pre-stored particle characteristics are matched with range measurements of robot at the stage of weight updating of particles. The real-time extraction of particle characteristics is avoided and the filter updating efficiency is improved. Through calculating the Kullback-Leibler distance between the discrete posterior probability distribution and the current particle set distribution, variable precision grid-particles are selected adaptively. The variable precision localization approach balances the filtering efficiency and the locating precision, and the "kidnapped robot" problem can be solved. Simulation results show that DSPF improves the locating efficiency while keeping the filtering accuracy.