nvidia-isaac/cuVSLAM

CUDA-accelerated visual odometry and SLAM library for fast, accurate, low-latency camera tracking

What it solves

cuVSLAM provides a high-performance solution for Visual Odometry and Simultaneous Localization and Mapping (SLAM). It allows robots and devices to track their position and map their environment in real-time using camera data, eliminating the need for manual parameter tuning through the use of engineered defaults that adapt to different environments and motion profiles.

How it works

The library leverages CUDA acceleration to process visual data efficiently. It supports multiple tracking modes based on the available sensor hardware:

  • Mono: Uses a single camera (scale-ambiguous).
  • RGBD: Uses one RGB-D camera with aligned depth.
  • Multicamera: Uses two or more cameras, including at least one stereo pair.
  • Inertial: Combines a stereo pair with an IMU for increased robustness.
  • Multisensor: An experimental mode that fuses any mix of RGB, RGB-D cameras, and an optional IMU using cuNLS (CUDA nonlinear least squares).

Who it’s for

It is designed for robotics developers and researchers working with NVIDIA hardware (Desktop, Server, or Jetson Orin/Thor) who need accurate, real-time visual tracking and mapping capabilities.

Highlights

  • CUDA-Accelerated: Optimized for high computational efficiency and real-time performance.
  • Zero-Tuning: Works out of the box with automatic adaptation to environment and motion.
  • Flexible Sensor Support: Compatible with various rigs, from single cameras to complex multi-sensor arrays.
  • Broad Integration: Provides C++ and Python APIs, as well as official ROS2 support via Isaac ROS.

Related

  • Project
    NVIDIA-ISAAC-ROS/isaac_ros_visual_slam

    A GPU-accelerated ROS 2 package for visual simultaneous localization and mapping (VSLAM) that provides high-precision odometry for robots using stereo cameras and IMUs.

  • Project
    ACFR-RPG/DynOSAM

    DynoSAM is a Stereo/RGB-D Visual Odometry pipeline for Dynamic SLAM that estimates camera poses and object motions while mapping static backgrounds.

  • Project
    stella-cv/stella_vslam

    stella_vslam is an open‑source visual SLAM library (monocular, stereo, RGB‑D) that supports many camera models, lets you save/load maps, and offers modular APIs plus ROS integration. It’s a community fork of OpenVSLAM, actively maintained under a BSD‑2 license.

  • Project
    ouguangjun/Leg-KILO

    A real-time LiDAR-based SLAM system designed for high-dynamic platforms like legged robots, featuring a two-stage ESKF frontend and factor graph backend for robust mapping.

  • Project
    luigifreda/pyslam

    A hybrid Python/C++ modular framework for Visual SLAM that integrates diverse feature extractors, depth prediction models, and semantic segmentation for trajectory estimation and 3D reconstruction.