Structure-Preserving Learning for Nonlinear Nonholonomic Dynamics

Learning the dynamics of mechanical systems subject to nonlinear nonholonomic constraints is challenging, as data-driven models may violate the underlying physical constraints, leading to inconsistent predictions. To address this problem, we develop a constraint-preserving learning framework based on the Lagrange-d'Alembert-Chetaev equations and tangency to the constraint submanifold, which together define an affine space of admissible accelerations. Using the kinetic-energy metric and the Moore-Penrose pseudoinverse, we construct a metric projector and the corresponding Chetaev correction. Embedding this geometric structure directly into a Gaussian process prior ensures that both prior and posterior dynamics satisfy the acceleration constraints exactly. We characterize the resulting reproducing kernel Hilbert space (RKHS), derive an adapted-coordinate representation, and establish nonexpansiveness and consistency transfer. Numerical experiments on a three-dimensional directional flocking system with up to 50 agents demonstrate exact constraint preservation and robustness to constraint-incompatible observations.

Authors

Publication Details

Journal
arXiv (Cornell University)
Published
2026-11-12
DOI
https://doi.org/10.48550/arxiv.2611.16796
Primary Topic
Gaussian Processes and Bayesian Inference
Type
preprint
Controls
|||
ALL TIME
JAN
FEB
MAR
APR
MAY
JUN
JUL
AUG
SEP
OCT
NOV
preprint

Structure-Preserving Learning for Nonlinear Nonholonomic Dynamics

Leonardo Jesus Colombo, Thomas L. Beckers, Anthony M. Bloch
arXiv (Cornell University)
Gaussian Processes and Bayesian Inference
preprint

Structure-Preserving Learning for Nonlinear Nonholonomic Dynamics

Leonardo Jesus Colombo, Thomas L. Beckers, Anthony M. Bloch
preprint en

Abstract

Learning the dynamics of mechanical systems subject to nonlinear nonholonomic constraints is challenging, as data-driven models may violate the underlying physical constraints, leading to inconsistent predictions. To address this problem, we develop a constraint-preserving learning framework based on the Lagrange-d'Alembert-Chetaev equations and tangency to the constraint submanifold, which together define an affine space of admissible accelerations. Using the kinetic-energy metric and the Moore-Penrose pseudoinverse, we construct a metric projector and the corresponding Chetaev correction. Embedding this geometric structure directly into a Gaussian process prior ensures that both prior and posterior dynamics satisfy the acceleration constraints exactly. We characterize the resulting reproducing kernel Hilbert space (RKHS), derive an adapted-coordinate representation, and establish nonexpansiveness and consistency transfer. Numerical experiments on a three-dimensional directional flocking system with up to 50 agents demonstrate exact constraint preservation and robustness to constraint-incompatible observations.

arXiv (Cornell University)
Gaussian Processes and Bayesian Inference
AI Navigator

Ask Laika to Summarize, Analyze, and Connect papers live on the map.

Summarize Papers & Methodologies

Extract key findings, datasets, and comparative methods across publications.

Benchmark Rankings & Visual Analytics

Rank top research institutions, authors, funders, topics, and journals by Field-Weighted Citation Impact (FWCI) and paper volume with instant charts.

Connect Distant Disciplines

Bridge topological clusters on the map to find hidden collaborative intersections.

Laika access coming soon