Skip to content

README.md

Latest commit

 

History

History
125 lines (89 loc) · 4.49 KB

README.md

File metadata and controls

125 lines (89 loc) · 4.49 KB

Jetson Example: Run NVBlox Mapping on NVIDIA Jetson

img

Isaac ROS NVBlox is a high-performance GPU-accelerated 3D mapping framework developed by NVIDIA for real-time robotic perception. Unlike monocular depth estimation models, NVBlox consumes true depth input from RGB-D cameras or stereo cameras to construct accurate 3D scene representations.This case enables you to quickly deploy the necessary environment for nvblox to run on your reComputer with just one click.

Detailed instructions for environment configuration can be found at:Deploy NVBlox with Orbbec Camera

Main process run it will:

  1. Download nvblox_images.tar from the built-in OneDrive share link into ~/.cache/jetson-examples/nvblox
  2. Run docker load -i on that archive
  3. Build the derived image and prepared host/container workspaces
  4. Launch the static Gemini2 NVBlox demo

Requirements

  • NVIDIA Jetson Orin
  • Ubuntu 22.04
  • JetPack 6.x
  • Docker with NVIDIA Container Runtime
  • Orbbec Gemini2 or another Orbbec camera that provides /camera/color/* and /camera/depth/*
  • Roughly 60GB free disk space for the cached archive, derived image, and managed workspace

Usage

Run the full prepare + demo flow:

cd jetson-example/
pip install .
reComputer run nvblox

Prepare only:

NVBLOX_MODE=prepare reComputer run nvblox

Run only after preparation:

NVBLOX_MODE=run reComputer run nvblox

Force a rebuild of the prepared host/container workspaces:

NVBLOX_FORCE_REBUILD=1 reComputer run nvblox

Run headless:

NVBLOX_HEADLESS=1 reComputer run nvblox

Override the managed workspace root:

MANAGED_ROOT=/path/to/nvblox_demo reComputer run nvblox

Override the built-in OneDrive archive settings:

NVBLOX_IMAGE_SHARE_URL='https://...'
NVBLOX_IMAGE_ARCHIVE_NAME='nvblox_images.tar'
NVBLOX_IMAGE_CACHE_DIR="$HOME/.cache/jetson-examples/nvblox"
reComputer run nvblox

Cleanup

reComputer clean nvblox

This removes the managed workspace, logs, partial downloads, the derived image local/isaac_ros_nvblox_orbbec:jp6-humble, and the running demo container if it exists.

It keeps:

  • the cached base archive in ~/.cache/jetson-examples/nvblox
  • the loaded base image imported from nvblox_images.tar

Troubleshooting

  • The default path checks ordinary Gemini2 color/depth readiness, not stereo IR capability.
  • Host readiness now requires only:
    • /camera/color/camera_info
    • /camera/depth/camera_info
    • /camera/color/image_raw
    • /camera/depth/image_raw
  • Container readiness now checks host camera discovery through /camera/color/camera_info and /camera/depth/camera_info.
  • The runtime success criterion is static map output from /nvblox_node/static_esdf_pointcloud or /nvblox_node/static_map_slice.
  • usb speed: 5000 Mbps is not treated as proof that the full demo is healthy. The final authority is whether host color/depth, container visibility, static TF, and static map output all succeed.
  • If the host driver exits and Gemini2 falls back to usb_present_no_video, the run path still attempts automatic recovery with udev refresh and USB rebind so you can usually retry without unplugging the camera.
  • If the run still fails, use the built-in connectivity debugger:
bash reComputer/scripts/nvblox/scripts/debug_runtime_connectivity.sh

That debug path follows the same stages as the default runtime:

  1. Gemini2 device state
  2. Host ROS discovery environment
  3. Container ROS discovery environment
  4. Host color/depth readiness
  5. Container camera visibility
  6. Managed static TF availability
  7. Static NVBlox output

Notes

  • This example does not use docker pull for the base image path.
  • The OneDrive downloader resolves the anonymous download.aspx?...tempauth=... URL from the preview page before downloading.
  • NVBLOX_MODE=run expects an already prepared MANAGED_ROOT.
  • The host camera is launched with ros2 launch orbbec_camera gemini2.launch.py publish_tf:=false tf_publish_rate:=0.0.
  • The container workspace now centers on nvblox_examples_bringup static Orbbec launches and removes the old default dependence on Visual SLAM.
  • The managed static TF chain is generated inside the prepared container workspace rather than relying on device-published TF.
  • Headless mode switches the default launch file to orbbec_debug.launch.py, while GUI mode uses orbbec_example.launch.py.