NVBlox voxel artifacts appear in low-light environment

Robotics & Edge Computing Robotics - Isaac Isaac ROS
,
1

Hello,
I am having some difficulty with what appear to be random voxel artifacts in the field of view of my depth cameras. These artifacts create single-voxel obstacles in the cost map near the robot and, when used with NAV2, cause it to randomly replan its route or prevent it from moving altogether; see the clip below. The “artifacts” are the darker voxels that keep appearing and disappearing above and away from the robot.

I suspect my test environment (nighttime, with bright artificial lights in an open space) may play a factor, but I can’t seem to find a combination of parameters that removes these artifacts. The closest thing was running nvblox in dynamic mode, but though there were fewer “artifacts” and they decayed faster, they still appeared.

I am running the realsense_example.launch.py launch file with the default nvblox_base.yaml parameters in release 3.2 of the isaac_ros_nvblox package.

I am using two Realsense cameras and the program is running on an AGX Orin Developer Kit with 64GB of RAM and a 1TB NVME. With the camera setup listed below, camera0 is also being used for VSLAM.

camera1 = Realsense D455f (at the front)
camera0 = Realsense D435 (at the rear)

Can I change a set of parameters to remove these, or is something else the issue? I’d also appreciate your help understanding what could have caused it. Although my indoor testing has been minimal, I haven’t experienced anything like this indoors.
Thanks in advance.

2

The random voxel artifacts you’re encountering in your depth cameras are likely caused by a combination of low-light conditions and artificial lighting, which can interfere with the accuracy of depth perception, especially with cameras like the Realsense. Adjusting camera settings such as exposure, gain, and depth quality may help reduce these issues. Additionally, fine-tuning NVBlox parameters like voxel size, decay rate, and update frequency can improve the mapping process in dynamic environments. Make sure both cameras are well-calibrated and synchronized to avoid misalignment, and consider testing in different lighting conditions or using a single camera setup to isolate the problem. Lastly, ensure your NAV2 and VSLAM configurations are optimized to handle the noise in the environment.

1 Like
3

Hi @joejoo

(and thank you also @pathofbuild for your reply)

I forwarded your issue to the engineering to have a more detailed reply. I keep you posted

Best,
Raffaello

1 Like
4

Thanks @pathofbuild for the suggestions. I will try them out.

@Raffaello, thanks. I’ll keep an eye out for their feedback as well.