MAC-VO
MAC-VO is an award-winning, learning-based stereo visual odometry system that won the ICRA 2025 Best Paper and Best Paper on Robot Perception awards. It introduces a metrics-aware covariance approach to improve pose estimation accuracy and robustness in robotic navigation. Key features include support for dense mapping without additional computational cost, a Fast Mode offering a 2x speedup to reach 12.5 FPS on 480x640 images via mixed-precision inference, and optimized pose graph optimization. The system provides official ROS-2 integration, downloadable real-world trajectory datasets captured with ZedX Stereo cameras, and comprehensive documentation for extending capabilities or using it as a boilerplate for custom visual odometry projects. MAC-VO requires CUDA Runtime 12.4 or higher, Python 3.10 or newer, and at least 6 GB of VRAM for standard operation, with Fast Mode reducing requirements to approximately 2.7 GB. The software is open-sourced under the MIT license and includes Docker support for easy deplo