Ask Your Question
0

How does MSF estimate heading ?

asked 2016-07-11 15:54:35 -0500

abhi gravatar image

Hi All,

I am relatively new to MSF and am trying to set it up for estimating the pose(absolute) of a ground vehicle. There are two update pose sensors (in addition to IMU for prediction step). One pose sensor provides absolute pose measurements in the world frame and the second one provides a relative pose (odometric data). The setup seems to work okay but the heading seems to be off. More precisely, the movement of the robot seems to be interpreted along the global frame and not the local frame. To clarify, a x velocity translates the robot along the global x axis and not the local x axis. I know this is not a lot to debug the problem, but is there something that I am doing wrong ? If some information/links could be provided as to the part where heading is estimated and how it is being estimated, it would be really helpful. Thanks.

-Abhishek

edit retag flag offensive close merge delete

Comments

It would help to have a little more information about your system - ROS distro, launch files, bag files etc. But first, check that your local and global frames have the same origin and orientation - if they don't the odometry can be very different in each one.

M@t gravatar image M@t  ( 2016-07-11 18:20:50 -0500 )edit

Distro: Indigo, launch file being used: position_pose_sensor.launch (supplied with MSF). The data is composed of: pose data from wheel encoders on a simulated robot and absolute pose data from vicon. The part about the global and local frames having same origin and orientation is satisfied.

abhi gravatar image abhi  ( 2016-07-11 18:49:25 -0500 )edit

1 Answer

Sort by ยป oldest newest most voted
1

answered 2016-07-11 22:13:56 -0500

M@t gravatar image

updated 2016-07-11 22:14:40 -0500

This isn't a proper answer sorry but it's too long an explanation for comments and it may give you some leads to look into.

  1. I would make sure ethzasl sensor fusion works on Indigo because from the wiki page it looks like Groovy is the latest distribution that's supported, so there may be compatibility issues (hopefully someone more familiar with ROS than me can tell you).
  2. I don't have any personal experience with ethzasl (I'm using the robot_localization package). But what I can say is that I see similar behavior quite a bit, and it's usually because sensor data is being converted to odometry in a reference frame, and somewhere along the way an angular offset is added to the data.

    For example if you line up your robot with the 'X' axis of your global frame and drive it in a straight line but your odometry in that frame shows the robot travelling at an angle to the 'X' axis. E.g:

    image description

    My global, local and GPS frames line up with the 'X' and 'Y' axis, but my global (red) and GPS (green) data is reported at an angle.

    Is this similar to what you're seeing? I suggest you try something similar, drive your robot in a straight line outside and plot the odometry in each frame in Excel.

    If this looks familar it's usually because your IMU heading has been distorted (nearby metal objects will do that) or because yaw-related parameter has been set incorrectly. Is one of your sensors a GPS? If so then it will almost certainly be doing some sort of geodetic conversion to produce an absolute pose - and for such conversions it's critical that parameters such as your magnetic declination and yaw offset are correct (as per REP-103 and REP-105, most IMU's read 0 when pointing North, whereas ROS expects them to read 0 pointing East). If you're using a provided launch file your robot may not report sensor data in the way the launch file expects it to.

I'm sorry I can't give you more specific advice, but hopefully this will give you some things to check. And in general, the more information you can provide about the problem the easier it is for others to debug - if you could post some bag files or graphs of your odometry similar to the one above that would help.

edit flag offensive delete link more

Your Answer

Please start posting anonymously - your entry will be published after you log in or create a new account.

Add Answer

Question Tools

2 followers

Stats

Asked: 2016-07-11 15:54:35 -0500

Seen: 199 times

Last updated: Jul 11 '16