Ask Your Question
0

Jackal odometry producing large covariances?

asked 2018-08-02 14:13:46 -0500

asabet gravatar image

updated 2018-08-02 14:54:23 -0500

I'm getting exploding covariance readings (along the diagonal) when echoing my /odometry/filtered topic on a jackal robot

header:    seq: 51936   stamp: 
    secs: 1533236107
    nsecs: 579915762   frame_id: "odom" child_frame_id: "base_link" pose:    pose: 
    position: 
      x: -1.0615022635
      y: 0.781039664382
      z: 0.0574243827245
    orientation: 
      x: -0.000565326105048
      y: 0.00129452020895
      z: -0.722981516966
      w: 0.690865928201   covariance: [52.22660058406395,
2.36797290728725e-06, 1.0344690435191654e-05, -1.412923508928883e-37, -5.323328075794997e-37, -3.0325742350059946e-33, 2.3679729072872455e-06, 52.226580153537775, 4.630394046333056e-06, 6.415510432550216e-39, -1.52209745995386e-35, 1.378174221195611e-34, 1.0344690435191593e-05, 4.630394046333093e-06, 62.66584761441557, 2.4284539729416056e-39, 6.071418793042712e-33, -5.009970111481826e-38, -1.4129235089288823e-37, 6.415510432550213e-39, 2.4284539729415573e-39, 9.999974359129776e-10, 3.7247437261868537e-28, -2.646392184397966e-26, -5.323328075794997e-37, -1.5220974599538599e-35, 6.071418793042711e-33, 3.7247437261815377e-28, 9.999974359129776e-10, -1.0056936786572167e-25, -3.0325742350059946e-33, 1.3781742211956115e-34, -5.009970111481829e-38, -2.6463921843979677e-26, -1.0056936786572173e-25, 9.999987179531678e-10] twist:    twist: 
    linear: 
      x: -1.36494277252e-16
      y: 0.0
      z: 0.0
    angular: 
      x: 0.00114722208953
      y: -0.000894948616335
      z: -0.000156898909114   covariance: [0.0015009578520837957,
0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0015009578520837957, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0020931884517782426, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.21732790378605e-07, 8.251167061986038e-27, -4.2753698250399015e-24, 0.0, 0.0, 0.0, 8.251167062032128e-27, 1.21732790378605e-07, -1.3789888806793583e-24, 0.0, 0.0, 0.0, -4.275369825039899e-24, -1.3789888806793574e-24, 1.2178973261379077e-07].

What are ways of mitigating this?

edit retag flag offensive close merge delete

Comments

Can you please repost the output using the 101010 button to enable formatting vs using the " button (which is meant for quoting things)? It's difficult to read the output as is.

jayess gravatar imagejayess ( 2018-08-02 14:31:53 -0500 )edit

Done. Thoughts?

asabet gravatar imageasabet ( 2018-08-02 14:54:34 -0500 )edit

1 Answer

Sort by ยป oldest newest most voted
2

answered 2018-08-02 15:31:21 -0500

stevejp gravatar image

updated 2018-08-02 15:32:19 -0500

So the /odometry/filtered topic is being published by an instance of robot_localization. It is likely set up to fuse wheel velocities and measurements from the internal IMU (orientation). This means the EKF has no absolute x/y/z measurements, so it is expected that the covariance would increase over time. You'd need to fuse absolute x/y/z measurements into the EKF if you didn't want the covariance to explode.

edit flag offensive delete link more

Comments

Okay, how do I do that? Are you saying to use wifi or GPS? Also, my covariances explodes with very little movement, especially for the angular measurements. It basically starts out extremely uncertain and remains that way.

asabet gravatar imageasabet ( 2018-08-02 15:36:36 -0500 )edit

You can also feed it a pose from some sort of localization node, e.g., amcl. You are integrating velocities, so you will integrate their errors, too.

When the robot rotates, the linear covariances will grow much faster once the robot starts translating again. This is normal.

Tom Moore gravatar imageTom Moore ( 2018-08-03 04:07:52 -0500 )edit

If you don't like the growth rate in the covariance, you can (a) decrease the covariance coming from your wheel encoders, and/or (b) decrease the process_noise_covariance for the robot's pose and velocity variables.

Tom Moore gravatar imageTom Moore ( 2018-08-03 04:13:17 -0500 )edit

Are you saying I can just arbitrarily set to the encoder covariances and process_noise-covariance to what whatever I want? How would this reflect the actual uncertainty from those nodes?

asabet gravatar imageasabet ( 2018-08-03 14:38:08 -0500 )edit

Yes, you can. And it depends on your robot and your application. I am not familiar with how the wheel odom covariance was generted for the Jackal, nor how they tuned the process noise, but I'd be willing to bet that you can adjust the latter until you feel the covariance growth is realistic.

Tom Moore gravatar imageTom Moore ( 2018-08-05 10:13:33 -0500 )edit

Your Answer

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

Add Answer

Question Tools

1 follower

Stats

Asked: 2018-08-02 14:13:46 -0500

Seen: 139 times

Last updated: Aug 02 '18