Grid-based Particle SLAM
Yiannis-Christos Psaradakis
Constantine Kalivas
SLAM is an acronym for "Simultaneous Localization and Mapping".
Grid-based SLAM, also known as FastSLAM 2.0, is a re-iteration of the original Landmark-based FastSLAM 1.0
This approach uses what is called an "occupancy grid", a 2D Map where each tile keeps a state if it's occupied or not. Every tile represents a small area of the map.
The localisation part of SLAM is taken care of by a Particle filter.
Each particle keeps it's own belief of the map state and pose of the robot.
Every particle updates it's position according to the result of the odometry, with a small deviation from the mean (calculated using a normal distribution)
Since we're using ROS, we are making use of the laser scan data provided by the simulation, without making any assumptions about errors.
Internally, we use a probability map, identical in size to the grid. Each tile keeps a probability of the tile being occupied. It is a mean of all previous measurements