rf2o have degraded performance in hallways

[HaoguangYang]

I have tested rf2o in hallways with width approx. 2m * 12m dimensions and it won't work. Though it works fine in feature-rich rooms, as soon as the robot enters the hallway it hardly recognizes the motion of robot, The wall at the end of the hallway will not help with localizing the robot. I suggest that its fitting process consider the distance of scanned points to the robot and weight it accordingly, since further points have less density and should take same weights as the points near-by. Also the global covariance matrix is 0, which makes it hard to integrate with other filtering algorithms such as EKF.

[tianb03]

This is so called corridor effect... You could only use a wheel-encoder to solve this problem but not any matching process

[JGMonroy]

Hi @HaoguangYang
Yes, what you describe is correct. As @tianb03 pointed out, scan matching algorithms are unable to estimate the odometry in such situations since the laser data does not contain valuable information.

The option you mention to increase the weight of distant points is dangerous. It is true that for this specific situations it can work, but in general, this is not true. Furthermore, laser errors are usually proportional to the distance measured, so far away points usually present higher noise levels.

[zlg9folira]

Any update on this topic? I am facing the same issue in hallways even with (very few) objects/doors which make a non-symmetric scan. I was trying to look around to see if there is any similar package that takes laser scans and publishes odometry and tf (odom -> base_link) - it seems all work the same way including srf_laser_odometry by @JGMonroy.

Wheel odometry drifts by the time (the same in my case) and degrades rf2o pose estimation. Any idea would be appreciated.