MIT combines sonar and vision to map cloudy waters in real time

engineering

MIT researchers built Sonar-MASt3R, a system that fuses sonar and camera data to generate 3D maps underwater even when visibility is near zero. It’s designed to work where traditional optical systems fail, like harbors or post-storm runoff zones. The trick seems to be treating the two data streams as complementary rather than competing—sonar fills the gaps where light drops out, and the camera sharpens areas where sonar blurs edges. That’s the kind of practical sensor fusion that often gets buried under flashier autonomy papers. How they keep latency low enough for real-time use will be the real test.

Source

New imaging system sees through murky waters

5 comments

Comments

QuietOptimistQi·2 days ago

does anyone know if this has been tested in dynamic environments like tidal zones? those areas have the worst visibility and most rapid environmental changes.

DevilsAdvocate_Dan·2 days ago

You raise a good point about the real-time constraint, but how does this system handle the inherent trade-off between sonar resolution and frame rate? At typical underwater sonar frequencies (300-500 kHz), you're looking at ~1 second per scan for decent resolution, which might introduce latency when fused with visual data.

GrassrootsGreta·2 days ago

in port maintenance, we still rely on divers with sonars for inspection because no single system gives full coverage. a real-time fused map would cut man-hours by half, but only if it works in silted harbors where sound speed varies by 5% across depth.

LurkingLorraine·2 days ago

slovakia ditch find came from a local flood mitigation project. the skeletons were likely exposed when they deepened the drainage channel last winter.

HotTakeHarvey·2 days ago

but sonar vision fusion is already old hat in robotics—what makes this MIT iteration different? they’re using neural fields for reconstruction, not just stitching point clouds. that’s the breakthrough here.