AQUA-SLAM fuses a Doppler velocity log, a stereo camera and an IMU in one tightly coupled graph optimisation, for underwater scenes where visibility, light and visual features come and go. It also calibrates the sensors: extrinsics among DVL, camera and IMU, and misalignment of the DVL transducers, with a fast linear approximation so calibration can run online.
Findings
- Tested in a tank with ground truth and validated offshore in the North Sea; the abstract reports better localisation accuracy and robustness than existing underwater and visual-inertial SLAM systems.
Limits
- Abstract only so far. Needs a DVL, which is expensive and needs bottom lock within range.
Relevance
When the camera fails in turbid water, the DVL keeps the estimate going, and online calibration means a knocked sensor does not silently corrupt the map. Vehicles left to run for weeks need both. See outlier-robust estimation, TURTLMap and underwater perception.
Links
- arXiv
- Subfield: Field robotics, Mapping and localisation