← sources

Share some of my recent work on RGB SLAM Camera

source: reddit (reddit:robotics)published 2026-11-19 02:08ingested 2026-11-20 02:39

retained text

I recently ran my RGB slam camera RoboBatonNeo on the ETH FPV dataset, currently sitting at 3rd place.

It's a little difficult. This dataset is pretty extreme:

☞ In the hardest sequence, speed reaches 100km/h

☞ Acceleration approaches 7G (69.85 m/s²), angular velocity up to about 20 rad/s

☞ Monocular images suffer from severe motion blur and pixel displacement at these speeds

Most of the top-ranked methods use stereo cameras + IMU. Our setup uses a single camera + IMU and runs on a board with a power consumption of 0.6 W.

Stacking high-speed dynamics + monocular + extremely low power makes it quite challenging to keep the algorithm both lightweight and stable. Logging this result for further iteration.

Anyone working on drone visual navigation or SLAM is welcome to chat.

I will share the work on Github once I've put everything together. Future updates will go there too.

Github: https://github.com/myrobotproject

document fetched 2026-11-20 00:39 · canonical https://reddit.com/r/robotics/comments/1xzxn1f/share_some_of_my_recent_work_on_rgb_slam_camera