Abstract—This paper presents an improved unscented fast simultaneous localization and mapping (UFastSLAM) using the geometric information of particles. The information are the pose of particles and their geometric relation, which are utilized in both the importance weight and the resampling steps. In the importance weight step, all particles are grouped via their weights and updated more accurately using the weight compensation scheme. In addition, the particle depletion problem is overcome by the particle formation technique using geometric relation between particles. The superior performance of the proposed approach over UFastSLAM is validated by the well-known Victoria Park dataset.

Index Terms—rao- blackwellized particle filter, UFastSLAM, EM algorithm, triangular mesh generation

Cite: Seung-Hwan Lee, Jung H. Oh, and Beom H. Lee, "Improved Unscented FastSLAM Using Geometric Information of Particles," International Journal of Mechanical Engineering and Robotics Research, Vol. 5, No. 1, pp. 43-46, January 2016. DOI: 10.18178/ijmerr.5.1.43-46