Surprise-Driven Exploration with Rao-Blackwellized Particle Filters for Efficiently Constructing Occupancy Grid Maps

  • Authors:

  • Youbo Cai;Masumi Ishikawa

  • Affiliations:

  • Department of Brain Science and Engineering, Kyusyu Institute of Technology, Kitakyusyu, Japan 808-0196;Department of Brain Science and Engineering, Kyusyu Institute of Technology, Kitakyusyu, Japan 808-0196

  • Venue:
  • ICONIP '09 Proceedings of the 16th International Conference on Neural Information Processing: Part II
  • Year:
  • 2009

Quantified Score

Hi-index 0.00

Visualization

Abstract

Proposed is a novel algorithm for surprise-driven exploration with Rao-Blackwellized Particle Filters (RBPF) for simultaneous localization, mapping, and exploration. "Exploration" is expected to find the optimal path for efficient simultaneous localization and mapping (SLAM). We propose to adopt the concept of surprise defined by Kullback-Leibler (KL) divergence between posterior belief and prior belief. During exploration, it evaluates each path by trading off its cost against the degree of surprise for the corresponding expected map in RBPF. We also propose to automatically generate candidate paths for evaluation instead of providing them from the outside. Simulation experiments demonstrate the effectiveness of the proposed algorithm. Compared with the previous studies that use information gain-driven exploration, the proposed method shows superior performance in selecting path closing, which drastically reduces uncertainty of robot poses and accordingly improves the accuracy of the resulting map.