DEV Community

Aman Sachan
Aman Sachan

Posted on

VoxelNav: Real-time Semantic Voxel Mapping for ROS2 at 100ms

#ros2 #robotics #mapping

I built VoxelNav because existing SLAM solutions are either too slow, too expensive, or too dumb.

The Problem

  • OpenGS-SLAM: 30s latency, needs A100 GPU
  • OctoMap: 500ms, no semantics
  • RTABMap: 1s, desktop only

Autonomous robots need real-time semantic understanding.

What VoxelNav Does

ROS2-native node that converts LiDAR/RGB-D data into semantic voxel grids:

  • 100ms end-to-end latency on Jetson Nano
  • Works with any ROS2-compatible sensors
  • Nav2-compatible costmap output

Quick Start 

cd ~/ros2_ws/src
git clone https://github.com/AmSach/voxelnav.git
colcon build --packages-select voxelnav
ros2 launch voxelnav voxelnav.launch.py
Enter fullscreen mode Exit fullscreen mode

Benchmarks (Jetson Nano)

Mode Latency Memory
Geometry only 30ms 50MB
Full semantic 100ms 150MB

GitHub: https://github.com/AmSach/voxelnav

Top comments (0)