initial pose estimate

[takahiko-hasegawa-hdjp]

Hi there.

I am currently using rtabmap (rgbd) to do map matching (visual localization). I can create a nice map using external wheel encoders but when in localization mode, it is very difficult to do map matching especially because i am doing localization in an outdoor environment (also with very low texture in the scene!).

I am moving the robot in a farm but struggling with initial position; tried 2D pose estimate which helps a lot but i have to do it every time. The starting position of the robot is always different so I cannot statically put the initial postion values.

Do you have any advice on doing initial position matching where there is not a lot features, no gps and cannot do 2d pose estimate?

Thanks!

[matlabbe]

For outdoor localization, you may want to use a feature more robust to lighting variations. See https://arxiv.org/pdf/2103.03827.pdf for a comparison of some of them that are available in rtabmap (in condition to build with right dependencies). For example, using SIFT or SURF features may help.