Hybrid State Estimation of Uncertain Nonlinear Dynamics Using Neural Processes

Fuente: arXiv
Saved in:
Bibliographic Details
Main Authors: Hunter, Devin, Enyioha, Chinwendu
Format: Preprint
Published: 2025
Subjects:
Online Access:
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866914039127343104
author Hunter, Devin
Enyioha, Chinwendu
author_facet Hunter, Devin
Enyioha, Chinwendu
contents Various neural network architectures are used in many of the state-of-the-art approaches for real-time nonlinear state estimation in dynamical systems. With the ever-increasing incorporation of these data-driven models into the estimation domain, models with reliable margins of error are required -- especially for safety-critical applications. This paper discusses a novel hybrid, data-driven state estimation approach based on the physics-informed attentive neural process (PI-AttNP), a model-informed extension of the attentive neural process (AttNP). We augment this estimation approach with the regression-based split conformal prediction (CP) framework to obtain quantified model uncertainty with probabilistic guarantees. After presenting the algorithm in a generic form, we validate its performance in the task of grey-box state estimation of a simulated under-actuated six-degree-of-freedom quadrotor with multimodal Gaussian sensor noise and several external perturbations typical to quadrotors. Further, we compare outcomes with state-of-the-art data-driven methods, which provide significant evidence of the physics-informed neural process as a viable novel approach for model-driven estimation.
format Preprint
id arxiv_https___arxiv_org_abs_2509_12522
institution arXiv
publishDate 2025
record_format arxiv
spellingShingle Hybrid State Estimation of Uncertain Nonlinear Dynamics Using Neural Processes
Hunter, Devin
Enyioha, Chinwendu
Systems and Control
Various neural network architectures are used in many of the state-of-the-art approaches for real-time nonlinear state estimation in dynamical systems. With the ever-increasing incorporation of these data-driven models into the estimation domain, models with reliable margins of error are required -- especially for safety-critical applications. This paper discusses a novel hybrid, data-driven state estimation approach based on the physics-informed attentive neural process (PI-AttNP), a model-informed extension of the attentive neural process (AttNP). We augment this estimation approach with the regression-based split conformal prediction (CP) framework to obtain quantified model uncertainty with probabilistic guarantees. After presenting the algorithm in a generic form, we validate its performance in the task of grey-box state estimation of a simulated under-actuated six-degree-of-freedom quadrotor with multimodal Gaussian sensor noise and several external perturbations typical to quadrotors. Further, we compare outcomes with state-of-the-art data-driven methods, which provide significant evidence of the physics-informed neural process as a viable novel approach for model-driven estimation.
title Hybrid State Estimation of Uncertain Nonlinear Dynamics Using Neural Processes
topic Systems and Control
url https://arxiv.org/abs/2509.12522