Certifiable optimization provides global optimality guarantees for non-convex state estimation in robotics, typically by solving convex semidefinite programming (SDP) relaxations with interior-point methods. A new arXiv paper (2610.20490) addresses what happens when physical sensing degeneracy becomes severe.
The abstract notes that under such degeneracy, the convex SDP relaxations solved via interior-point methods are the subject of concern—the sentence is cut off in the announcement, so the specific claim is not available. The paper is listed as new, and its title indicates a focus on observability degeneracy.
Given the truncated abstract, the full contribution is not yet clear. The work appears to aim at making certifiable estimation robust when observability is degenerate, a scenario that can occur in real-world robotic systems.