ZED 2i outdoor positional tracking path drifting problem

Software
1

Hi Stereolabs team,

I am using a ZED 2i camera with a ZED Box Mini, running SDK 5.3.0, for an outdoor quadruped robot mapping and localisation project. The previous connection problem has now been solved, and the ZED 2i can run outdoors properly. I recorded and ran the positional tracking during an outdoor robot test (refer to the csv file). The recorded data includes position, quaternion rotation, timestamp, and tracking state.

I converted the ZED position data into a 2D top-down pathway (see the HTML document) and overlaid it on a Google Maps screenshot to compare the ZED trajectory with the actual robot route. The tracking state was reported as OK during the recording. However, the generated ZED path appears rotated or misaligned compared with the real robot movement path.

I’ve modified the positional tracking sample to:

1. set the camera to HD720 @ 60 fps,

2. enable pose smoothing,

3. enable Area Memory (save/load map.area),

4. log pose (XYZ + quaternion + timestamp) to CSV.

I have attached images that show:

1. The actual robot pathway drawn on QGIS Maps.

2. The ZED-generated pathway overlaid on the QGIS Maps.

From the comparison, the ZED path shape is visible, but the orientation does not align correctly with the real movement direction. This looks like an IMU or orientation alignment issue that I observed in a previous test. Thus, I would like help to solve the ZED 2i camera position drift problem.

Thank you very much.

Best regards,

Hong Wei

Comparison of ZED trajectory with real-world coordinates
Comparison of ZED trajectory with real-world coordinates1280×720 250 KB

zed_3d_position_rotation_interactive_map.html (8.3 MB)

zed_position_log_19700101_011002.csv (2.7 MB)

2

Hi Hong Wei,

Thank you for the report and the CSV log. We went through your data and would like to ask a few questions before drawing any conclusions.

From your log we noted the following: the path forms a closed loop of about 294 m over 16 minutes, the trajectory ends 9.3 m from its starting point, the vertical coordinate rises steadily to about +8.5 m, and there are a few sudden position corrections (0.7 to 2.3 m) when the robot comes back near its start area.

Could you help us with the following?

  1. What does the project ultimately need: geo-referenced positions (GPS/UTM) for map overlay, or a locally consistent trajectory for navigation? What accuracy are you targeting?

  2. Is a GNSS receiver available on the robot, or could one be added?

  3. Which positional tracking mode do you use (PositionalTrackingParameters.mode: GEN_1, GEN_2 or GEN_3)? Ideally paste your full InitParameters and PositionalTrackingParameters.

  4. How did you align the ZED trajectory with the map before overlaying it? What did you use as the heading or north reference?

  5. Is the test site flat, or does the route climb around 8 m in elevation? Did the robot return to its exact physical start point?

  6. How is the camera mounted on the robot (height, pitch, rigidly fixed or vibration-damped)?

  7. Could you attach the map overlay screenshot, and tell us whether the angular offset looks constant along the run or grows over time?

  8. Could you run ZED_Diagnostic on the ZED Box Mini and attach the report?

With these answers we can identify what is happening and recommend the right setup for your project.

Best regards,

Tristan