Total Speech Recognition icon Total Speech Recognition — Offline Windows speech recognition with open dictionary, confidence filtering, and scriptable phrase actions for ARC robots Try it →
Asked
1 Participant
D435i NMS offset and floor cropping

D435i NMS Offset And Floor Cropping

Running ARC on a DF Robot LattePanda (Windows) inside a small 2WD base, with an Intel RealSense D435i plugged into the USB 3.0 port. I’m using the Intel Realsense D435i robot skill to feed Navigation Messaging System data for obstacle mapping. The camera is mounted 120 mm forward of the robot’s center of rotation, at 280 mm height, with about a 5 downward tilt. I select the device by serial, hit START, and it streams reliably (30 fps, ~40% CPU on the LattePanda). However, in my NMS map, obstacles show up about 120-150 mm closer than tape-measure reality, and if the Image Align bottom bar is too low I get a "phantom wall" ~0.25 m from the bot due to the floor.

What I’ve tried: in the skill’s Configuration I set Mounted Radius Offset to -120 mm (since the tooltip says negative is toward the front), but flipping to +120 mm seems to shift the plotted returns the other way without actually fixing the absolute distance mismatch. I also adjusted the Image Align top/bottom bars; raising the bottom eliminates the floor but then I lose low obstacles like a 100-150 mm tall pet bowl. I confirmed the D435i is on a powered USB 3.0 hub to rule out power drops, and I’m viewing ARC over WiFi remote desktop to avoid bandwidth spikes.

What is the correct sign convention and reference point for Mounted Radius Offset in the Intel Realsense D435i skill (center-of-rotation to the camera’s optical center or the front face), and how should I set Image Align for a 280 mm height and 5 tilt so I avoid floor returns without losing low obstacles?



Related Hardware (view all EZB hardware)
Lattepanda by DF Robot
LattePanda with Synthiam ARCx runs headless, loads EZB firmware on the built-in Arduino, and controls robots from any browser
Wi-Fi / USB
Servos 12
Camera
Audio
UART 1
I2C
ADC 5
Digital 12

Related Robot Skill (view all robot skills)
Intel RealSense T265 SLAM tracking for ARC way-point navigation with pose streaming, mapping, and low-CPU visual odometry.

ARC Pro

Upgrade to ARC Pro

With Synthiam ARC Pro, you're not just programming a robot; you're shaping the future of automation, one innovative idea at a time.

Author Avatar
PRO
Canada
Twitter YouTube GitHub
#1  

You’re on the right track, but there’s a better way to think about this in ARC:

1) Mounted Radius Offset is not "front face to wall"

In the Intel RealSense D435i skill, Mounted Radius Offset is the distance from the robot’s center of rotation to the camera’s sensing origin along the robot’s forward axis.

  • Negative = camera is in front of the robot center
  • Positive = camera is behind the robot center

So for your robot, -120 mm is the correct starting value.

Important detail

Use the camera’s optical center / sensing origin, not the plastic front face of the camera.
If you measured from the front of the camera housing, that can easily explain the 120-150 mm mismatch.

Why your plot may still look off

If the absolute error is still around 120-150 mm after using -120 mm, then one of these is likely true:

  • you measured from the wrong physical reference point
  • the robot center of rotation is not actually where you think it is
  • the D435i is tilted downward, which changes where obstacle points land in the map
  • the floor is being interpreted as a valid obstacle plane

2) The floor issue is usually solved by vertical filtering, not by changing radius offset

The Image Align top/bottom bars are effectively a height filter for what part of the depth image gets used.

If the bottom bar is too low, the floor enters the point cloud and ARC sees a "phantom wall" nearby.

If you raise it too much, you’ll lose low obstacles like bowls, feet, and curb edges.


3) Best practice for your mounting height and tilt

With:

  • camera height: 280 mm
  • down tilt: ~5
  • small 2WD base

Start with this approach:

  1. Set Mounted Radius Offset to -120 mm
  2. Raise the bottom crop just enough so the floor disappears
  3. Lower the top crop only if needed to keep the detection band wide enough for low objects

Practical rule

You want the crop to include:

  • the floor-adjacent region where low obstacles exist
  • but exclude the part of the image that is mostly floor reflections / ground plane

With a 280 mm mount height and only 5 downward tilt, the floor will usually start dominating the lower part of the depth frame fairly close to the camera. So the bottom crop needs to be set slightly above the horizon where the floor first appears, not all the way down.


4) What I would do on your robot

Try this sequence:

A. Verify the offset first

  • Keep Mounted Radius Offset = -120 mm
  • Test with a known obstacle at a measured distance
  • If returns are consistently off by the same amount, adjust in small steps:
    • -100 mm
    • -130 mm
    • -150 mm

Only change by 10-20 mm at a time.

B. Tune floor rejection

  • Increase the bottom crop until the phantom wall disappears
  • Then lower it slightly until a low object, like your bowl, returns again

This is a balancing act. The exact value depends on:

  • your camera pitch
  • how high the camera is mounted
  • the floor color / reflectivity
  • whether the robot is level when driving

5) If you want the most reliable mapping, use the sensor skill’s physical mounting data

For navigation, ARC works best when the skill’s physical parameters match the robot exactly.
If the camera is tilted even a little, the depth projection can shift enough to make map distances look wrong.

If the D435i skill does not expose a separate pitch/tilt compensation control, then the proven workaround is:

  • mount the camera as level as possible
  • use the image crop to remove the floor
  • keep the offset to the actual center-of-rotation-to-camera-origin measurement

6) My recommendation for your exact setup

For a first clean setup:

  • Mounted Radius Offset: -120 mm
  • Image Align bottom bar: high enough to remove floor returns
  • Image Align top bar: only as high as needed to preserve low obstacle detection
  • keep the camera as close to level as possible, or reduce the tilt if you can

7) If you want, I can help you dial it in precisely

If you share:

  1. a screenshot of the D435i skill configuration
  2. a screenshot of the NMS map with the phantom wall
  3. the exact measurement method you used for the 120 mm offset

I can help you determine whether you should keep -120 mm or move it slightly, and estimate a better crop band for your 280 mm mounting height.

Also, if your bigger goal is accurate low-height obstacle detection for navigation, let me know what navigation skill you’re using on top of NMS, because the best tuning depends on that too.

For code-related tuning or automation, the ARC AI Assistant can also help you rewrite scripts or debug project logic directly in ARC:
https://synthiam.com/Products/AI-Assistant

I am a customer support AI and still learning. Please verify important answers in our support documentation. For advanced AI features, use the AI Script Agent built into ARC.