NVIDIA-ISAAC-ROS/isaac_ros_nvblox

What it solves

Isaac ROS Nvblox provides a way for robots to build a 3D map of their environment in real-time. It solves the problem of creating accurate 3D scene reconstructions and 2D costmaps for navigation, allowing robots to avoid obstacles based on vision data from depth cameras or 3D LiDAR.

How it works

The system processes depth images, color images, and pose data (often from a SLAM node) to reconstruct a 3D scene on the GPU. This 3D reconstruction is then "sliced" into a 2D costmap that can be used by navigation frameworks like Nav2 for path planning.

It supports three modes of operation:

1. Static: Assumes the environment is unchanging.
2. People Reconstruction: Uses a DNN model (PeopleSemSegNet) to create segmentation masks for people, separating them from the static environment.
3. Dynamic Reconstruction: Handles more general dynamic objects in the scene.

Who it’s for

Robotics developers building autonomous navigation systems that require real-time 3D mapping and obstacle avoidance using NVIDIA GPUs.

Highlights

• GPU Accelerated: Uses the nvblox C++ library for high-performance 3D reconstruction.
• Vision-Based Navigation: Integrates directly with Nav2 via a costmap plugin.
• Dynamic Scene Support: Specialized modes for mapping people and other moving objects.
• Real-time Visualization: Streams mesh updates to RViz for live 3D scene monitoring.