Paperlane

no. 16 · Tuesday 17 November 2026

On Covariance Steering with Nonlinear Measurements

Daniel C. Qi and Kenshiro Oguri

math.OC

Abstract

This paper examines two ways nonlinear measurements complicate the statistical predictions during covariance steering. The covariance steering problem optimizes a control policy across a distribution of possible trajectories rather than a single specific trajectory. This approach has attracted growing interest as a way to account for uncertainty in aerospace applications such as trajectory optimization, stationkeeping, and landing. In many of these applications, covariance steering can account for measurement uncertainty through output feedback. Because standard formulations assume linear dynamics and measurements, applications with nonlinear measurements typically use an extended Kalman filter to estimate the state. This paper shows that techniques used to improve filter convergence, such as measurement underweighting, can reduce the accuracy of conventional covariance steering predictions by violating a key assumption. This is addressed with a new convex formulation that relaxes the assumption. This paper also demonstrates that adjusting the control policy during execution can improve agreement between predicted and actual state statistics. A numerical example of angles-only spacecraft rendezvous illustrates the benefits of the proposed method, and discussions are made regarding the implications of this generalized covariance steering formulation.

fields
Optimization and Control
pdf ↗ page arXiv:2611.20337 ↗