DAVIO combines Depth Anything 3 and IMU for real-time dense metric SLAM, code released

zhenjun_zhao · x · 2026-09-24

DAVIO is a dense monocular-inertial SLAM system pairing the feed-forward depth model Depth Anything 3 with a VIO filter. A five-image window plus preintegrated IMU forms a feature-free linear system for early start-up; metric filter poses then condition the depth model, correcting residual scale only along viewing rays. On EuRoC it starts markedly earlier and localizes more accurately than prior mappers, and degrades far less on building-scale sequences with real odometry. Code is open-sourced.

Original post →

More from Embodied

Embodied channel →