Moving Horizon Estimation for Robust State Tracking

This direction is merged into another direction: moving horizon estimation. Its last version stays readable and citable here.

Version 1 of 121 papers30 conceptsfrozen 2026-10-04

In control theory and robotics, knowing the current state of a system—its position, temperature, or chemical concentration—is rarely as simple as reading a sensor. Sensors are noisy, some variables are hidden, and physical systems must obey strict laws. While the Kalman Filter has been the industry standard for decades, it often struggles when systems become highly nonlinear or when physical constraints (like "concentration cannot be negative") are violated.

Moving Horizon Estimation (MHE) offers a more robust alternative by treating state estimation not as a recursive update, but as an optimization problem solved over a sliding window of recent history.

The Intuition of Hindsight

Traditional recursive filters, such as the Extended Kalman Filter (EKF), are "forgetful." They process the latest measurement, update their internal state, and discard the raw data. If a sensor provides a wildly incorrect outlier, the filter has no way to "look back" and realize that the measurement contradicted the trajectory established over the last ten seconds.

As Hong Jang, Jay H Lee, and Su-Hang Choi explain in A Survey on State Estimation of Nonlinear Systems, MHE maintains a "moving horizon"—a fixed-length buffer of the most recent measurements and inputs. Instead of asking "What is the state now given the last state?", MHE asks: "What is the most likely sequence of states that explains this entire window of data?"

By optimizing over a trajectory rather than a single point, MHE gains a form of "temporal inertia." This makes it significantly more resilient to outliers and signal noise, a benefit highlighted by Keck Voon Ling and Kun Cao in their work on Neural Ranging Correction. In "urban canyons" where GPS signals bounce off buildings, MHE uses historical context to maintain a stable trajectory even when individual sensor readings are severely degraded.

Enforcing the Laws of Physics

The most significant advantage of MHE is its ability to handle hard constraints. In many real-world applications, we know certain states are physically impossible. For example, in retinal laser treatment, Viktoria Kleyman and Karl Worthmann use MHE to ensure that estimated tissue temperatures and laser absorption parameters remain within realistic physical bounds.

Recursive filters often produce estimates that drift into "impossible" territory because they lack a mechanism to enforce inequality constraints. Because MHE is formulated as a constrained optimization problem, it can explicitly forbid these impossible values, leading to superior stability in complex, multi-parameter scenarios.

From Power Grids to Physiology

The versatility of MHE allows it to solve problems across vastly different scales:

  • Renewable Energy: Milan Vukov and Moritz Diehl demonstrated the first real-time MHE scheme for wind turbines, allowing turbines to maximize power capture while respecting structural fatigue limits.
  • Power Systems: In Moving-horizon dynamic power system state estimation, Gang Wang and Seung-Jun Kim use semidefinite relaxation to turn non-convex grid estimation into a convex problem, outperforming traditional filters in tracking fast-moving dynamics.
  • Medicine: Seth Siriya and Matthias A. Müller applied MHE to thyroid hormone tracking, showing that the windowed approach can reconstruct hidden physiological variables even when blood tests are irregular or doses are forgotten.

Estimation Without a Model

Traditionally, MHE requires a precise physics-based model of the system. However, recent research has moved toward Data-driven MHE (Dd-MHE). This approach allows for state estimation without an explicit mathematical model, instead using historical trajectories to represent system dynamics.

Tobias M. Wolff, Matthias A. Müller, and Victor G. Lopez have developed frameworks for linear discrete-time systems that bypass system identification entirely. Similarly, Haitao Chang and Panfeng Huang applied data-driven MHE to track the spin of "noncooperative" space debris, where a physics model is often unavailable.

By combining the rigorous constraints of optimization with the flexibility of a sliding window, MHE provides a bridge between classical control and modern, data-rich environments, ensuring that our estimates of the world remain both accurate and physically grounded.

Papers in this version

History & citationVersion 1 · 2026-10-04

Merged into another direction · Permanent link

View frozen version 1

Cite this direction

Latest version (a living page):

@misc{lacuna_moving_horizon_estimation_for_robust_sta_a6bbb474,
  title = {Moving Horizon Estimation for Robust State Tracking},
  author = {{Lacuna}},
  howpublished = {Lacuna research direction},
  year = {2026},
  url = {https://lacuna.tiptreesystems.com/direction/moving-horizon-estimation-for-robust-state-tracking/rdr_a6bbb474e0ec4d0c8d88b37a9da030b3},
  note = {Latest version of a living research direction; cite a version for a fixed text. Accessed 2026-11-19},
}

This exact version (version 1, frozen text):

@misc{lacuna_moving_horizon_estimation_for_robust_sta_a6bbb474_v1,
  title = {Moving Horizon Estimation for Robust State Tracking},
  author = {{Lacuna}},
  howpublished = {Lacuna research direction},
  year = {2026},
  url = {https://lacuna.tiptreesystems.com/direction/moving-horizon-estimation-for-robust-state-tracking/rdr_a6bbb474e0ec4d0c8d88b37a9da030b3/v/1},
  note = {Version 1, frozen 2026-10-04. Accessed 2026-11-19},
}