Skip to main content

Documentation Index

Fetch the complete documentation index at: https://mintlify.com/UZ-SLAMLab/ORB_SLAM3/llms.txt

Use this file to discover all available pages before exploring further.

ORB-SLAM3 provides a comprehensive set of example programs for testing and running SLAM with various sensor configurations and datasets. All examples are built automatically when you compile the library.

​
Available Example Programs

The library includes examples for all supported sensor configurations:

Monocular

Single camera SLAM
  • mono_euroc
  • mono_tum_vi
  • mono_kitti
  • mono_realsense_D435i
  • mono_realsense_t265

Stereo

Dual camera SLAM
  • stereo_euroc
  • stereo_tum_vi
  • stereo_kitti
  • stereo_realsense_D435i
  • stereo_realsense_t265

RGB-D

Depth camera SLAM
  • rgbd_tum
  • rgbd_realsense_D435i

Visual-Inertial

Camera + IMU fusion
  • mono_inertial_euroc
  • mono_inertial_tum_vi
  • stereo_inertial_euroc
  • stereo_inertial_tum_vi
  • mono_inertial_realsense_D435i
  • mono_inertial_realsense_t265
  • stereo_inertial_realsense_D435i
  • stereo_inertial_realsense_t265

​
Supported Datasets

ORB-SLAM3 has been extensively tested on several public datasets:

​
EuRoC MAV Dataset

Recorded with two pinhole cameras and an IMU. Contains 11 sequences with ground truth trajectories.
  • Visual (monocular/stereo)
  • Visual-Inertial (monocular/stereo + IMU)
  • Pinhole camera model
Learn more about EuRoC examples →

​
TUM-VI Dataset

Recorded with two fisheye cameras and an IMU. Includes indoor and outdoor sequences.
  • Visual (monocular/stereo)
  • Visual-Inertial (monocular/stereo + IMU)
  • Fisheye camera model
Learn more about TUM-VI examples →

​
KITTI Dataset

Autonomous driving dataset with stereo cameras.
  • Monocular sequences
  • Stereo sequences
  • Large-scale outdoor environments
Learn more about KITTI examples →

​
Live Camera Support

Run SLAM in real-time with Intel RealSense cameras.
  • RealSense D435i (stereo infrared + IMU)
  • RealSense T265 (fisheye stereo + IMU)
Learn more about RealSense examples →

​
Running Examples

1

Build the library

First, ensure ORB-SLAM3 is compiled:
cd ORB_SLAM3
./build.sh
This creates executables in the Examples/ folder.
2

Download or prepare data

For dataset examples, download the sequence data. For live cameras, connect your device via USB.
3

Run the example

Execute the appropriate program with required arguments:
./Examples/<Type>/<executable> Vocabulary/ORBvoc.txt <settings.yaml> <data_path>

​
Expected Outputs

When running examples, ORB-SLAM3 produces:
A Pangolin viewer window displays:
  • Current frame with tracked features
  • 3D map points
  • Camera trajectory
  • Keyframes
Saved trajectories in TUM or EuRoC format:
  • CameraTrajectory.txt - All frame poses
  • KeyFrameTrajectory.txt - Keyframe poses only
Format (TUM):
timestamp tx ty tz qx qy qz qw
Terminal displays:
  • Tracking status (SLAM, Tracking Only, Lost)
  • Frame processing times
  • Number of tracked features
  • Map statistics
When compiled with REGISTER_TIMES:
  • ExecTimeMean.txt - Execution time statistics
  • Per-module timing breakdowns

​
Evaluation Tools

ORB-SLAM3 includes scripts for trajectory evaluation:

​
EuRoC Evaluation

./euroc_eval_examples
Processes all EuRoC sequences and computes RMS ATE (Absolute Trajectory Error) against ground truth.

​
TUM-VI Evaluation

./tum_vi_eval_examples
Evaluates drift at sequence end points where ground truth is available.

​
Python Evaluation Tools

The library uses Python scripts for trajectory alignment: 
python evaluation/evaluate_ate_scale.py <groundtruth.txt> <estimated.txt>
Evaluation scripts require Python with NumPy installed.

​
Common Arguments

Most examples follow this pattern: 
./<executable> <vocabulary> <settings> <sequence_path> [output_name]
ArgumentDescription
vocabularyPath to ORB vocabulary file (usually Vocabulary/ORBvoc.txt)
settingsYAML configuration file with camera calibration and parameters
sequence_pathPath to dataset sequence folder
output_name(Optional) Prefix for output trajectory files

​
Performance Tips

Hardware

  • Use a powerful CPU (i7 or better recommended)
  • Run on native Linux for best performance
  • Close unnecessary applications

Configuration

  • Adjust ORBextractor.nFeatures for speed/accuracy trade-off
  • Use image downsampling for faster processing
  • Disable viewer for maximum speed

​
Next Steps

EuRoC Dataset

Run examples with pinhole cameras and IMU

TUM-VI Dataset

Run examples with fisheye cameras

KITTI Dataset

Process autonomous driving sequences

RealSense Cameras

Real-time SLAM with live cameras

Build docs developers (and LLMs) love

Get started for freeTalk to us