Estimation-Aware Trajectory Optimization with Set-Valued Measurement Uncertainties
Aditya Deole and Mehran Mesbahi · 2025
Read Estimation-Aware Trajectory Optimization with Set-Valued Measurement Uncertainties by Aditya Deole and Mehran Mesbahi (2025) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2501-09192.txt.
Introduction
Aditya Deole and Mehran Mesbahi present a 2025 paper on planning flight paths that keep a spacecraft easy to track. When sensors report fuzzy, set-valued measurements, some trajectories leave the craft's state badly estimated. Their framework folds estimation quality directly into the path design, using enveloping ellipsoids to bound the unknown measurement spread.
For well-behaved sensor maps, the authors define an observability score that is concave along candidate paths, which makes maximizing it tractable. Planning routines then optimize the score for systems with locally linearized dynamics. Test cases cover vision-based navigation scenes. The flagship example sends one satellite to rendezvous with an uncooperative target, guided by an onboard machine learning estimator. Paths chosen this way keep the chaser informed, not just on course.
Catalog record
| CREATOR | Aditya Deole and Mehran Mesbahi |
|---|---|
| YEAR | 2025 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 103 KB |
| VERSION | 1.0 |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2501-09192
Browse the full library · About the Tower of Babel
Full text
The opening of the work, from the library's hosted text:
Estimation-Aware Trajectory Optimization with Set-Valued Measurement Uncertainties Aditya Deole and Mehran Mesbahi Abstract In this paper, an optimization-based framework for generating estimation-aware trajectories is presented. In this setup, measurement (output) uncertainties are state-dependent and set-valued. Enveloping ellipsoids are employed to characterize state-dependent uncertainties with unknown distributions. The concept of regularity for set-valued output maps is then introduced, facilitating the formulation of the estimation-aware trajectory generation problem. Specifically, it is demonstrated that for output-regular maps, one can utilize a set-valued observability measure that is concave with respect to the finite horizon state trajectories. By maximizing this measure, estimation-aware trajectories can then be synthesized for a broad class of systems. Trajectory planning routines are also examined in this work, by which the observability measure is optimized for systems with locally linearized dynamics. To illustrate the effectiveness of the proposed approach, representative examples in the context of trajectory planning with vision-based estimation are presented. Moreover, the paper presents estimation-aware planning for an uncooperative Target-Rendezvous problem, where an Ego-satellite employs an onboard machine learning (ML)-based estimation module to realize the rendezvous trajectory. Nomenclature \bm{x}_{k}\in\mathbb{R}^{n_{x}},\bm{u}_{k}\in\mathbb{R}^{n_{u}},\bm{y}_{k}\in\mathbb{R}^{n_{y}} = discrete time state, input, and output triplets at time k \mathbb{S}^{+} = positive definite matrices (dimension implicit) \mathcal{E}(\bm{x},Q) = ellipsoid centered at \bm{x} characterized by Q\in\mathbb{S}^{+} \mathcal{Y}_{\bm{x}} = output set corresponding to state \bm{x} \mathcal{X}\subset\mathbb{R}^{n_{x}},\mathcal{U}\subset\mathbb{R}^{n_{u}} = sets of feasible states and inputs \Lambda(\mathcal{Y}_{\bm{x}}) = function defining maximum size of the set \mathcal{Y}_{\bm{x}} \lVert.\rVert = 2-norm A,B,C = state space system matrices d(.,.),\;d_{s}(.,.),\;d_{\Gamma}(.,.) = metrics for Euclidean, set, and tube distances \bm{x}_{0:T},\;\bm{u}_{0:T} = discrete state and input sequences from time zero to T \Gamma_{\bm{x}_{0}}\bm{u}_{0:T-1} = output tube mapped from initial state \bm{x}_{0} with an input sequence \bm{u}_{0:T} Y_{\bm{x}_{0:T}} = output tube generated for state sequence from time zero to T D_{\mathcal{O}}(Y_{{\bm{x}}_{0:T}}),\;D_{\mathcal{O}}^{\ell}(Y_{{\bm{x}}_{0:T}}) = Degree of observability of an output tube and its lower bound 1 INTRODUCTION T RAJECTORY planning often involves solution of optimization problems that efficiently guide a vehicle to a desired final state, while ensuring feasibility with respect to constraints on the vehicle’s state and input [ 1 ] . This work addresses a trajectory planning problem that accounts for the vehicle’s state estimation process—a consideration particularly critical in safety-critical systems where the accuracy of state estimation is essential. Traditionally, in offline planning, the estimation process is assumed to be independent of state and control, causing the synthesized trajectory to potentially traverse regions with high measurement uncertainty. Meanwhile, in many practical applications, large measurement errors or disturbances adversely affect the quality of the corresponding state estimation; in such scenarios, estimation-aware trajectory design becomes indispensable [ 2 ] . In order to follow a trajectory, an agent often employs a state estimator, mapping the process measurements to full state estimates, provided that the system is observable [ 3 , 4 , 5 , 6 , 7 ] . The performance of this estimator is closely tied to the quality of the sensor measurements. If the sensor performance is state-dependent, then visiting states where the sensor has better performance can improve the corresponding state estimation process. One common approach to design optimal trajectories for improved state estimation is to incorporate a secondary objective into the original trajectory optimization problem [ 8 , 9 ] . This augmented objective function must account for the quality of the estimation process with respect to errors in the measurement. Given knowledge of the measurement error distribution, a suitable metric for capturing the estimation quality can then be developed and embedded in the design objective. Performance metrics for evaluating the estimation performance include measures on error probabilities [ 10 ] , mean squared errors [ 11 , 12 , 13 ] , Fisher information [ 14 , 15 , 16 ] , and entropy and other metrics on the belief state. For example, a commonly used construct in settings with a Gaussian measurement noise is the Fisher Information Matrix (FIM) [ 15 , 16 ] . In a nutshell, FIM quantifies the information content about an unknown state (or parameter) in a measurement. In fact in the matrix case, the inverse of FIM provides a lower bound on the expected error covariance, using the Cramer-Rao inequality [ 17 , 18 ] , i.e., \mathbb{E}\{(\hat{\bm{x}}-\bm{x})(\hat{\bm{x}}-\bm{x})^{T}\}\geq{\cal I}^{-1} , where {\cal I} denotes the FIM, and \bm{x} and \hat{\bm{x}} denote the state of interest and its estimate, respectively; this lower bound represents the error covariance achieved by an unbiased estimator. Various properties of FIM, such its determinant, spectrum, and trace, have been used in different applications [ 19 ] . The FIM-induced metrics utilized for quantifying the information content of a measurement scheme are effective for parameter estimation problems when the underlying state is static. On the other hand, measures such as posterior FIM have been studied to analyze state estimation for linear time-invariant (LTI) systems [ 20 , 21 ] . Optimization-based approaches have also been developed to demonstrate the utility of error covariance metrics for trajectory planning problems [ 22 , 23 ] . Another framework for designing estimation-aware trajectories is via observability-based metrics. Specifically, the Observability Gramian (OG) of a system, mapping initial conditions to an output trajectory sequence, can be used to assess how “sensitive” state estimation is to perturbations in output measurements. In this direction, prior studies have demonstrated that properties of the Empirical Gramian (EG) can be leveraged to improve conditioning for the estimation process [ 24 , 25 ] . These works have shown that using EG generated from the model prediction, an optimization problem can be formulated to synthesize a qualitatively more “observable” trajectory. These trajectories have improved estimation performance based on the type of metric used on the EG [ 8 , 26 , 27 , 28 ] . These studies primarily focus on applications with deterministic output maps with Gaussian noise measurements, that can be propagated forward, multiple roll-outs at a time [ 29 , 30 ] . In addition, probabilistic models using covariance-based optimization have been examined in [ 31 , 32 ] . Other motion planning approaches use Rapidly Exploring Random Trees to determine estimation-aware trajectories [ 33 ] . And lastly, there are methods proposed in the literature that rely on application-specific correlations between estimation and the underlying state for trajectory design [ 34 ] . In this work, an alternative–yet complementary–perspective to the aforementioned lines of work has been provided, particularly for scenarios where measurements are set-valued and conventional methods for EG computation are infeasible due to Gaussian noise assumption. However, similar to metrics used EG-based approaches, here, a set distance metric is used to improve the conditioning of the estimation process over a trajectory. In the area of filtering for set-valued uncertainty, set-membership estimation has been investigated solely for the purpose of filtering, where estimation procedures for set-based outputs are analyzed [ 35 , 36 , 37 ] . To the best of our knowledge, studies on estimation-aware planning with set-based output uncertainties have been rather limited, signifying the need for “filtering” methods for scenarios where noise distributions are unknown yet bounded. This work specifically focuses on measurement spaces with state-dependent uncertainties that are often non-Gaussian, for example, as they arise in perception-aware planning. In this direction, an explicit approach for such planning problems