Best Practices for Quick Automatic Localization

Posted by dschnabel on
URL: http://official-rtab-map-forum.206.s1.nabble.com/Best-Practices-for-Quick-Automatic-Localization-tp6881.html

What is a good way for an autonomous robot to localize itself, knowing where in the map it is?

Background: Let's say I have a pre-recorded map and my robot starts up rtabmap not knowing where on the map it is located. In most cases it won't be able to localize itself right away. Right now I need to teleoperate the robot and drive around a bit before it will know where on the map it is located.

1. Would it make sense to start with a low Vis/MinInliers (e.g. 10) until localization was successful and then bump it up to something higher (e.g. 50) to get better accuracy? Any other options to play around with to help with quick localization?

2. At what pattern should the robot start driving around to find out where it is? Does a star pattern make sense (move 2m in one direction, then back to starting point and move 1m in a different direction)?

3. How does rtabmap indicate whether or not localization was successful? This would be helpful for the navigation stack to know when it's time for path planning.