Overview
This open source project is a full visual-inertial mapping and localization stack: stereo + IMU odometry online, fixed-lag optimization in the loop, AprilTag-based global corrections, and an offline pipeline to clean trajectories up and turn logged sessions into usable maps.
The system was implemented as a means to keep learning and immersing myself in all the things robotics. I made this in the after hours of my weekdays and through many weekends. It was hard to work on this bit by bit, sometimes only 30 minutes a day. It took longer than I anticipated. There’s just so much tooling one needs to develop around the main product, to debug, to visualize. I think this is to a point where it’s already worth sharing but there’s just so much to keep working on.
I’d recommend to check the linked Github repo to dive deeper. It contains multiple readmes that explain the most important parts of the system. I’ll try to keep this portfolio post somewhat technical, but talk also about the story, results and learnings.
I'm currently writing this post, It'll be ready mid June. In the meantime, feel free to check the Github repo.