Issues with D435i and D455 in nvblox and VSLAM: Odom Freezes Intermittently

Robotics & Edge Computing Robotics - Isaac Isaac ROS
1

Hello @Raffaello

I’m using both a D435i and a D455 to drive nvblox and VSLAM, but I’m encountering an issue with the VSLAM odometry. After running for a short while, the odometry process freezes, and the /odom topic stops publishing any data. After a few moments, the process seems to restart, but then it freezes again, and this cycle repeats.

image
image1440×1310 211 KB

image
image892×406 82.3 KB

image
image1602×650 105 KB

image
image1920×765 274 KB

image
image1920×158 64.8 KB

Why is there such a message output? It seems abnormal.

Could this be related to frame synchronization issues between the two cameras? If so, could you provide some suggestions or best practices for handling multi-camera setups using Intel RealSense cameras? Any advice or solutions to ensure proper synchronization would be greatly appreciated.

Thank you!

2

@Raffaello

I have another question regarding using RealSense cameras with ESS DNN Stereo. If I use the RealSense camera directly for processing, I notice I achieve a higher depth frame rate. However, I’m wondering if the realsense_splitter_node is necessary in this setup. Could I bypass it, or are there any specific advantages to using it with ESS DNN Stereo?

Thank you for your insights!

3

Hi @alex.jaywill

Thank you for your post.
What hardware are you using? Messages like these may indicate insufficient memory.

I’m looking the Isaac ROS ESS launch file and only use realsense2_camera::RealSenseNodeFactory isaac_ros_dnn_stereo_depth/isaac_ros_ess/launch/isaac_ros_ess_realsense.launch.py at main · NVIDIA-ISAAC-ROS/isaac_ros_dnn_stereo_depth · GitHub

If I don’t say something wrong, is not using the realsense_splitter_node.

Let me know.

Best,
Raffaello

5

Hi @Raffaello

I’m using Jetson Orin AGX 64GB hardware with two RealSense cameras. I have achieved hardware time synchronization between the two cameras, and the results in vSLAM are working very well.

If I want to use DNN depth information with nvblox, do I still need realsense_splitter_node? I’m asking because I noticed that emitter_enabled is set to true in the RealSense example in nvblox.