LiDAR-inertial-visual odometry combines geometric depth from LiDAR with visual measurements to estimate motion, but the sensors can fail independently. LiDAR struggles in under-constrained geometry, while vision fails under poor illumination or low texture. These degenerate conditions can degrade the fused estimate.
The new SA-LIVO system, described in a recent arXiv preprint, explicitly addresses this problem with a subspace-aware degeneracy handling approach. Rather than treating all sensor failures equally, the method identifies the specific subspace in which the system is degenerate and adjusts the fusion accordingly.
This is a practical step toward more reliable odometry in challenging real-world environments, where no single sensor is always trustworthy. The paper is available on arXiv under identifier 2606.25699.