Researchers from the National University of Singapore (NUS) have developed a breakthrough in underwater robotics navigation with GeVI-SLAM, a gravity-enhanced stereo visual-inertial simultaneous localization and mapping (VI-SLAM) system. This innovation addresses long-standing challenges in underwater robotics, where traditional VI-SLAM systems often falter due to visual ambiguity and limited inertial sensor data.
GeVI-SLAM leverages the unique capabilities of stereo cameras to directly estimate depth, eliminating the need for scale estimation during inertial measurement unit (IMU) initialization. This advancement enables stable operation even in low-acceleration environments, a common scenario in underwater exploration. By precisely initializing gravity, the system decouples pitch and roll from pose estimation, solving a 4 degrees of freedom (DOF) Perspective-n-Point (PnP) problem. This approach significantly reduces computational time and improves outlier rejection within a Random Sample Consensus framework.
The researchers further refined the system with a bias-eliminated 4-DOF PnP estimator, ensuring that relative pose accuracy improves as more features are detected. To handle dynamic motion, GeVI-SLAM refines the full 6-DOF pose while jointly estimating IMU covariance, allowing for adaptive weighting of the gravity prior. This ensures robust performance even in challenging underwater conditions.
Extensive experiments conducted on both simulated and real-world data demonstrate that GeVI-SLAM outperforms state-of-the-art methods in accuracy and stability. This breakthrough could revolutionize underwater robotics, enabling more precise navigation and mapping in environments where traditional systems struggle. The practical applications are vast, from deep-sea exploration to underwater infrastructure inspection, offering a more reliable tool for scientists and engineers working in these demanding conditions. Read the original research paper here.