Does size matter when doing Gmapping with Kinect?

edit

The accuracy of SLAM mainly depends on the range of the sensor, size of the environment compared to sensor Field of view (FoV) and uncertainty of motion and measurements. Generally, Kinect sensor is a low FOV (57 degrees) and short-range sensor (max 5.5m).

Gmapping builds locally consistence map using scan matching and builds a globally consistent map by fusing Rao-blackwellized particle filter. It needs some features ( objects) in the FoV to build a consistent map. If enough features are not available in sensor FoV for few frames, the local estimation will not give a good accuracy.

Also, the noise of the sensor is a quadratic function with the range of measurements. This also leads the wrong estimation of local maps. Therefore, Kinect can provide a promising result with Gmapping for small-scale environments.

You can use a higher number of particles to get more accurate maps. But, very large number of particles also may lead to diverge the estimation, because, there a higher probability to pick particles with low weights at resampling step.