Announcement

Collapse
No announcement yet.

Vuforia - accuracy of location

Collapse
X
  • Filter
  • Time
  • Show
Clear All
new posts

  • Vuforia - accuracy of location

    Our team is evaluating the accuracy of the location reported by Vuforia to see if we can use it reliably for autonomous.

    We are using the sample code from SDK : ConceptVuforiaNavRoverRuckus

    with following camera setting:
    ================================================== ================================================== ======== final int CAMERA_FORWARD_DISPLACEMENT = 0; // eg: Camera is ON the robot's center line final int CAMERA_VERTICAL_DISPLACEMENT = 135; // eg: Camera is 135 mm above ground when phone is placed on the ground final int CAMERA_LEFT_DISPLACEMENT = 0; // eg: Camera is ON the robot's center line ================================================== ================================================== ======== We placed the phone on the field portrait mode with back camera facing the target image. We tried the position exactly one tile in front of each image and center to the image. We noticed that the value for the axle that stick out of the image is always very accurate at 46 in for all four pictures. The heading is very accurate too. The z axle value is fixed around 5in though we expect it to be 0 as we set the camera vertical displacement as 135mm (the camera height to the bottom of the phone in portrait mode). The value for the axle which is parallel to the long side of the target pictures is not 0 either and stayed about 5 in too for all four target pictures. To get zero, we have to shift to the right of the picture for about 5in. We double checked that the target pictures are placed at the correct place on the border. What is causing this? The above lines are the only lines that we changed in the sample code. Thanks.

    FTC8565

  • #2
    Originally posted by FTC8565 View Post
    Our team is evaluating the accuracy of the location reported by Vuforia to see if we can use it reliably for autonomous.

    We are using the sample code from SDK : ConceptVuforiaNavRoverRuckus

    with following camera setting:
    ================================================== ================================================== ======== final int CAMERA_FORWARD_DISPLACEMENT = 0; // eg: Camera is ON the robot's center line final int CAMERA_VERTICAL_DISPLACEMENT = 135; // eg: Camera is 135 mm above ground when phone is placed on the ground final int CAMERA_LEFT_DISPLACEMENT = 0; // eg: Camera is ON the robot's center line ================================================== ================================================== ======== We placed the phone on the field portrait mode with back camera facing the target image. We tried the position exactly one tile in front of each image and center to the image. We noticed that the value for the axle that stick out of the image is always very accurate at 46 in for all four pictures. The heading is very accurate too. The z axle value is fixed around 5in though we expect it to be 0 as we set the camera vertical displacement as 135mm (the camera height to the bottom of the phone in portrait mode). The value for the axle which is parallel to the long side of the target pictures is not 0 either and stayed about 5 in too for all four target pictures. To get zero, we have to shift to the right of the picture for about 5in. We double checked that the target pictures are placed at the correct place on the border. What is causing this? The above lines are the only lines that we changed in the sample code. Thanks.

    FTC8565
    Are you using the Version 2 of the Vuforia targets (found at https://www.firstinspires.org/resour...-season-info)? There was an error in the initial files released that were in error on one axis.
    Technical Coach, Newton Busters FTC 10138, FLL 3077
    Chicago 'burbs, IL

    Comment


    • #3
      A couple of things to consider (in addition to using the version 2 targets):

      1) The op mode you're using is set up for landscape mode, not portrait mode (you'd need to change phone rotation to YZX {-90,0,90} for portrait);
      2) The position of the camera on the robot refers to the position of the camera lens, not the bottom, top, or sides of the phone.

      Comment


      • #4
        Thanks for all your suggestions. We did use the version 2 targets. And the position of the camera is the lens to the robot. To be precise, it is 125mm from the lens to the bottom of the phone instead of 135mm. The mode suggestion is the right one which fixed our problem. We didn't read the comment carefully to notice that the phone starts with top heading left of the robot therefore on Y axle. After we changed to YZX {-90,0,90}, we got accurate location reading now. Thanks. We will explore how to use the location info in autonomous.

        FTC8565

        Comment


        • #5
          We used the Vuforia for location last year to ensure exact placement for lowering an arm with a color sensor between the two balls. It worked very well at that distance (balancing board to pictograph). I wonder if the accuracy diminishes as distance to the pictograph increases. The angles would diminish which leads me to believe that accuracy diminishes as the distance increases.

          In practice, we made sure the pictographs were placed precisely 3 inches from the center of the stand for the jewels. However, at competitions, the pictographs were placed in an approximate area instead of an exact placement. The pictographs last year were changed on each round and the location would vary from round to round. The 1 inch variance part of the field setup guide was more than the variance we could have on the placement of the color sensor. Our arm knocked off a ball when it came down once. I knew immediately that the placement of the pictograph was wrong. Indeed, the field wall was about an inch from the edge of the mat, causing the pictograph to be an inch further over than it should have been.

          This year, the pictographs do not move. They will be placed in the beginning and remain for the day. However, if the field is slightly off or that placement is not precise, it could affect your reading. Just keep in mind that the field setup guide allows for a variance.

          Comment


          • #6
            Thanks for the information. Looks like Vuforia is accurate if within 1-2 tiles from the target picture and if the target picture is placed exactly as the spec. It is true that there are variances from tournament to tournament on how the picture is placed. We will have to take that into account for the strategy and may need another source to help.

            Comment

            Working...
            X