cuVSLAM not initializing

Omniverse Isaac Sim
, , , ,
1

Hello,

I am new to using Isaac Sim and ROS2. I am trying to implement SLAM for a research project I’m working on. I want to get a SLAM system running using the example scene before porting it to my own project. I have been following this tutorial, and I believe I installed everything properly but I haven’t been able to get the SLAM module running without errors. I have been trying to solve this for a while now but am getting pretty lost, the cameras on carter robot are succesfully publishing to Rviz2 but it seems like the SLAM module doesn’t recognize them. I included the exact error messages and some info on my system below.

Another detail that I’m not sure if its important or not is that my simulation seems to be running at only 10 fps, I have tried to increase this by lowering the viewport resolution and deactivating the lidar and “owl” cameras on the carter but the fps does not go up and the error still occurs. It says that there is still ~6gb or GPU memory available but doesn’t use any of it. Any help or guidance that you can provide would be very appreciated.

Thanks!

Isaac Sim Version

4.2.0

Operating System

Ubuntu 22.04

GPU Information

  • Model: NVIDIA GeForce RTX 4080 Laptop GPU,
  • Driver Version: 535.183.01
  • CUDA Version: 12.2

Error Messages

[component_container-1] [INFO] [1739164806.427901605] [visual_slam_node]: Initializing cuVSLAM.
[component_container-1] [INFO] [1739164806.428101732] [visual_slam_node]: Use use_gpu: true
[component_container-1] [INFO] [1739164806.428124195] [visual_slam_node]: Enable IMU Fusion: false
[component_container-1] [INFO] [1739164806.428145270] [visual_slam_node]: Time taken by CUVSLAM_CreateTracker(): 0.000004
[component_container-1] [ERROR] [1739164806.428169439] [visual_slam_node]: Failed to initialize CUVSLAM tracker: 4
[component_container-1] [INFO] [1739164806.428211073] [visual_slam_node]: Initializing cuVSLAM.
[component_container-1] [INFO] [1739164806.428249189] [visual_slam_node]: Use use_gpu: true
[component_container-1] [INFO] [1739164806.428261351] [visual_slam_node]: Enable IMU Fusion: false
[component_container-1] [INFO] [1739164806.428273345] [visual_slam_node]: Time taken by CUVSLAM_CreateTracker(): 0.000001
[component_container-1] [ERROR] [1739164806.428289145] [visual_slam_node]: Failed to initialize CUVSLAM tracker: 4

2

Please check if Failed to initialize CUVSLAM tracker: 4 is helpful.

4

Hey @kylerwitvoet! Could you please check which branch are you on for isaac_ros_common package? You need to make sure that you are on release-3.2 branch.
For isaac_ros_visual_slam, did you install debian package or did you build from source? If you built from source, please also make sure that you are on release-3.2 branch and source it after building it.

Failed to initialize CUVSLAM tracker: 4 is just a generic error code and it doesn’t really tell much information about your failure.

5

Hi @zhengwang @kylerwitvoet !

I confirm I have the same issue Failed to initialize CUVSLAM tracker: 4 - x86_64 - ZED Camera, with both ROSBag and ZED Camera tutorials

6

tried both from source and debian package

7

@alberto.remus What is your output of nvcc --version?

8

The issue persists. Can we have someone to help out. I tried on two different machines with CUDA 12.6 and continue to have see “Failed to initialize CUVSLAM tracker: 4” . In fact i see the same error when i try running with your rosbag

9

Hi @formegham could you please try changing the verbosity level of isaac_ros_visual_slam (link to ROS params) to 1 and see what is the error in the log file?

10

We will have a minor release for Isaac ROS 3.2.6 which will make the verbosity level of isaac_ros_visual_slam to 1 by default.

11

I was able to resolve the error by upgrading by Nvidia drivers to 570 which now has CUDA 12.8.
after that I was getting warnings of jitter threshold being high which I resolved by setting the image_jitter_threshold based on my camera fps
30 fps → threshold of 35 ms
10 fps → threshold of ~100 or 150 ms(additional buffer)

1 Like