Inaccuracy in 2D map navigation.

I'm using a differential drive mobile robot with ROS Indigo with the navigation parameters tuned fairly well.

Yet there is noticeable offset when it comes to the robot reaching the exact goal location. Whenever a 2D goal is given for it to navigate and reach within a map, the offset or undershoot i.e. the actual location which the robot reaches has a difference of ~0.3 meters, which i think is pretty big.

This is a major accuracy hit.

Even in in-place rotation with quaternion, there's undershoot with the orientation.

I understand that wheel slips, structure indeed influence motor odometry which impacts accuracy, but a offset this high would be bad.

Kindly suggest how this can be overcome. What changes would I have to include ?

Thank you!

Asked by spiritninja on 2019-09-11 06:56:55 UTC

Comments

Answers

See the parameters you have available to play with. Amongst them you'll find yaw_goal_tolerance and xy_goal_tolerance. Beware, making them too small will have the undesirable effect of localization / slippage causing overshoot and the robot will attempt to keep reaching the goal to your tolerance.

Asked by stevemacenski on 2019-09-11 07:52:20 UTC

Comments

Yeah that's true. I have adjusted it accordingly after performing multiple tests. I even went through the ROS navigation tuning guide and set my tolerance parameters accordingly. Yet the offset persists.

Asked by spiritninja on 2019-09-11 07:55:08 UTC

it might be localization issue in that case, the robot pose jumps temporary within the goal tolerances. In order for us to get more info please post an RViz image or video together with your planner config params

Asked by pavel92 on 2019-09-11 08:16:03 UTC

Also try to measure the literal distance, not an estimate. Print out when its stopping the exact distance, if you see it hovering really close to a clean number like 0.3 or a param default/set then maybe something's not getting propogated. If its random then that would point to localizer updates potentially.

Asked by stevemacenski on 2019-09-11 08:22:34 UTC