OpenVINS

The reference open platform for visual-inertial state estimation research

License

GPL-3.0

Overview

OpenVINS, from the Robot Perception and Navigation Group (RPNG) at the University of Delaware, is an open platform for visual-inertial navigation research. At its core is a filter-based estimator built on an on-manifold sliding-window Kalman filter (MSCKF lineage) with online camera-IMU calibration, and it ships with the simulators, evaluation tooling and documentation that made it the standard baseline in the VIO literature.

OpenVINS fuses camera images and IMU measurements into a real-time 6-DOF state estimate. The estimator is a type-based on-manifold extended Kalman filter in the MSCKF tradition: SLAM landmarks and MSCKF feature tracks share a sliding window, with first-estimates Jacobians for consistency, online calibration of camera intrinsics, extrinsics and time offset, and support for monocular, stereo and multi-camera setups. Around the estimator the project provides a full research ecosystem: a groundtruth-driven simulator, the ov_eval toolkit for trajectory-error evaluation, ROS 1 and ROS 2 wrappers, and detailed documentation of the underlying derivations. Its modular 'type' system lets researchers add new state variables (calibration parameters, new sensors) without rewriting filter internals, which is exactly why so many VIO papers benchmark against or build on OpenVINS.

Why it matters

OpenVINS is the standard open baseline for visual-inertial odometry: when new VIO methods are published, OpenVINS is the system they compare against, and its documented derivations plus evaluation tooling turned rigorous VIO research from a closed-lab specialty into something any group can reproduce.

Problem it solves

GPS-denied robots — drones indoors, AR headsets, legged and mobile robots — need accurate real-time pose from cheap sensors. OpenVINS provides a consistent, calibrated camera+IMU state estimator plus the simulation and evaluation infrastructure to trust and extend it.

Key takeaways

• On-manifold sliding-window filter (MSCKF lineage) with first-estimates Jacobians for consistency
• Online calibration of camera intrinsics, camera-IMU extrinsics and time offset
• Monocular, stereo and multi-camera support with ROS 1 and ROS 2 wrappers
• Ships a groundtruth simulator and the ov_eval trajectory evaluation toolkit
• The de-facto published baseline for visual-inertial odometry research

Who it is for

• VIO and state-estimation researchers
• Drone and AR/VR engineers needing GPS-denied localization
• Graduate students studying Kalman-filter-based estimation
• Robotics teams evaluating camera+IMU sensor stacks

Limitations

• Filter-based design trades some accuracy versus optimization-based systems on long trajectories
• No built-in loop closing in the core estimator (secondary ov_secondary module exists)
• Requires careful IMU noise characterization for best performance
• GPL-3.0 license limits closed-source productization