FastSLAM Using Compressed Occupancy Grids

Abstract

Robotic vehicles working in unknown environments require the ability to determine their location while learning about obstacles located around them. In this paper a method of solving the SLAM problem that makes use of compressed occupancy grids is presented. The presented approach is an extension of the FastSLAM algorithm which stores a compressed form of the occupancy grid to reduce the amount of memory required to store the set of occupancy grids maintained by the particle filter. The performance of the algorithm is presented using experimental results obtained using a small inexpensive ground vehicle equipped with LiDAR, compass, and downward facing camera that provides the vehicle with visual odometry measurements. The presented results demonstrate that although with our approach the occupancy grid maintained by each particle uses only40%of the data needed to store the uncompressed occupancy grid, we can still achieve almost identical results to the approach where each particle filter stores the full occupancy grid.

Distributed Monocular SLAM for Indoor Map Building

“…Most importantly FastSLAM supported nonlinear process models and nonGaussian pose distributions. In more recent work, FastSLAM by Cain and Leonessa [4] uses a compressed occupancy grid to reduce the data usage of each particle by 40%. Pei et al [5] used distributed unscented particle filter to avoid reconfiguring the entire system during vehicle state estimation.…”

Robust Visual Localization with Dynamic Uncertainty Management in Omnidirectional SLAM

“…Originally, laser [3][4][5] and sonar [6,7] were widely acknowledged. Despite the fact that these sensors are still strongly accepted nowadays [8][9][10], they are being slightly relegated as secondary sensory data to the detriment of visual sensors, which have been experienced an growing use recently. Digital cameras have had a great success, being one of the most feasible tools for acquiring information from the environment.…”