View my account

T265 accuracy in a confined area

Comments

3 comments

  • MartyG

    At the T265's launch, an Intel representative stated that the accuracy of its SLAM is approximately that of GPS at a 100 cubic meter range.

    In RealSense SDK version 2.31.0, its release notes say that "Relocalization [is] significantly improved for areas up to 50 sq. m: testing indicates fast (< 5 seconds) relocalization for typical environments and usage with non-static T265 device".

    If possible collisions with objects during the robot's travel are anticipated, the T265 can be paired with a depth camera on the robot for obstacle avoidance.  The Intel presentation in the YouTube link below illustrates this.  The part about depth camera pairing starts at 17 minutes 15 seconds into the video.

    https://youtube.com/watch?v=62vm0_RZ1nU 

    You should be able to import and export maps with the RealSense Viewer program. The subject is discussed in the link below.

    https://github.com/IntelRealSense/librealsense/issues/5068 

    0
    Comment actions Permalink
  • Pmazumder3927

    Is this true for a previously seen area? If so then GPS accuracy is very bad. I would expect for it to be much more accurate on a premapped area, and want to see if anyone has measured the error for this yet. I am looking for less than a foot of inaccuracy.

    0
    Comment actions Permalink
  • MartyG

    For detailed expert advice on the T265, I recommend asking at the RealSense GitHub forum by visiting the link below and clicking the New Issue button.  This is also the best place to ask for input from other T265 users, as that is where the majority of T265 users gather.

    https://github.com/IntelRealSense/librealsense/issues 

    In general though, the T265 cases that I have seen where error amounts have been reported indicate an error of around + or - 0.5 meters when 7 meters is travelled.  The discussion in the link below is an excellent resource for insights into tackling tracking drift.

    https://github.com/IntelRealSense/librealsense/issues/3970 

    0
    Comment actions Permalink

Please sign in to leave a comment.