Dead reckoning with a MiR100 about mir_robot

[mintar]

You can use tf to get the transform from the "map" frame to the "base_footprint" frame. This is the result of the AMCL localization (i.e., the sensor fusion from laser scanners, IMU and odometry). This is what the MiR uses internally as well. You don't need to do any fusion with the /odom topic yourself, since it's already included.

[rsr152]

Thanks, @mintar. I will give that a shot.
Really appreciate your help!

[rsr152]

I started exploring the use of tf but wanted to get more input on whether it would help at all with better estimation of robot pose in a situation where I want to use dead reckoning to get to a target pose (using only encoder feedback, no camera feedback).

I am currently extracting the current pose of the MiR using an API call and guide it to its target pose (assuming a target pose such as a waypoint has been defined earlier). All these coordinates (x, y, theta) and (xtarget, ytarget, thetatarget) are in the "map" frame. If the robot is "suffering" from encoder errors, I feel using tf will not help overcome these. There will be a need to capture feedback from other sensors like an overhead camera, or the onboard camera (which I have not leveraged yet).

Am I missing something? The encoder errors are something we have to live with, right?
Is there any way to adjust for it? Appreciate your thoughts...

[mintar]

As I said, the tf pose (which should be the same that you get from the MiR via an API call) already contains a sensor fusion between odometry (= encoders), IMU and laser scanners. If you have a normal amount of encoder slip and IMU drift, this is usually corrected by referencing the laser scans to the recorded map. The onboard camera is not useful for localization, since it is pointed downwards to the floor. If you get a lot of localization errors, I would recommend re-recording the map; this usually helps. If you have a highly dynamic environment, where the map changes drastically all the time, you might have to integrate an external positioning system.

[rsr152]

Thanks a lot for the suggestions. This is working well for now!

[stevemartinov]

@mintar I think you need to read the question a bit more careful before replying with the same comment as you replied before.

@rsr152 Yes, that is how AMRs perform the last meter drive in. You first use AMCL to navigate to some pose then when you have arrived, you can turn off the AMCL and slowly give commands directly to the motor controllers to move forward/backward X meters. 1 meter is nothing for wheel encoders and you will probably get mm accuracy