Simultaneous localization and mapping (SLAM) is related to robotic navigation involving optimal path planning and exploring unknown environments. The problems for SLAM increase with navigation in different environments and finding the optimal path for robotic localization. This paper presents solutions to SLAM and path planning.
SLAM and localization are briefly introduced, and the authors assert the advantage of using evidential grid mapping over probabilistic grid mapping. They present an evidential FastSLAM algorithm to solve SLAM and path planning problems, which comprises four steps: prediction, importance weighting, map update, and resampling. The authors further describe value iteration, importance weight, and inverse sensor model algorithms.
The authors employ odometry data from two SLAM datasets recorded by Cyrill Stachniss in the Cartesium building and by Dirk Hähnel in the Intel Research Lab. For robotic navigation, they use “the Cartesium environment ... [of] size 55m×30m ... consist[ing] of ca. 660,000 grid cells.” The authors make a quantitative comparison of evidential FastSLAM, pignistic transformation, and FastSLAM algorithms using “the ground truth information of the environment.”
Clemens et al. claim four contributions of their paper: they approximate “the joint distribution over the map and the robot’s pose using a Rao-Blackwellized particle filter”; “they provide a general solution based on Markov decision processes [... about] the robot’s behavior”; they “derive evidential forward and inverse models for range sensors”; and they show that “the evidential approach provides advantages both for SLAM and for navigation.”
The expected audience of this paper could be students and professionals working in the area of SLAM involving robotic or automated navigation. The authors point out the limitation of probability occupancy grid maps in being unable to distinguish between different types of uncertainty, which makes this paper an interesting read.