Abstract:In numerous robotics and mechanical engineering applications, among others, data is often constrained on smooth manifolds due to the presence of rotational degrees of freedom. Common datadriven and learning-based methods such as neural ordinary differential equations (ODEs), however, typically fail to satisfy these manifold constraints and perform poorly for these applications. To address this shortcoming, in this paper we study a class of neural ordinary differential equations that, by design, leave a given manifold invariant, and characterize their properties by leveraging the controllability properties of control affine systems. In particular, using a result due to Agrachev and Caponigro on approximating diffeomorphisms with flows of feedback control systems, we show that any map that can be represented as the flow of a manifold-constrained dynamical system can also be approximated using the flow of manifold-constrained neural ODE, whenever a certain controllability condition is satisfied. Additionally, we show that this universal approximation property holds when the neural ODE has limited width in each layer, thus leveraging the depth of network instead for approximation. We verify our theoretical findings using numerical experiments on PyTorch for the manifolds S2 and the 3-dimensional orthogonal group SO(3), which are model manifolds for mechanical systems such as spacecrafts and satellites. We also compare the performance of the manifold invariant neural ODE with classical neural ODEs that ignore the manifold invariant properties and show the superiority of our approach in terms of accuracy and sample complexity.

On the Existence of Linear Observed Systems on Manifolds with Connection

A class of fully nonlinear equations arising from conformal geometry

Topological equivalence canonical forms for linear multivariable systems without control

One-input linear control systems on the homogeneous spaces of the Heisenberg group -- The singular case

A Note on the Extended Kalman Filter on a Manifold

Observability is Sufficient for the Design of Globally Exponentially Convergent State Observers for State-affine Nonlinear Systems

Transforming A Nonlinear System To A P-Normal Form Via Coordinates Transformation

On Linearization of Mechanical Control Systems

On Geometric Realization of the General Manakov System

One-Dimensional Solution Families of Nonlinear Systems Characterized by Scalar Functions on Riemannian Manifolds

Regularization of covariance matrices on Riemannian manifolds using linear systems

A Reinvestigation On The Controllability And Observability Of Networked Dynamic Systems

Machine learning based state observer for discrete time systems evolving on Lie groups

Learning on Manifolds: Universal Approximations Properties using Geometric Controllability Conditions for Neural ODEs

A New Approach to Synchronization Analysis of Linearly Coupled Map Lattices

On the irreducibility and convergence of a class of nonsmooth nonlinear state-space models on manifolds

Determining Lorentzian manifold from non-linear wave observation at a single point

Log-linear Dynamic Inversion Control with Provable Safety Guarantees in Lie Groups

Observability of Linear Time-Invariant Systems with Relative Measurements: A Geometric Approach

Admissible transformations and Lie symmetries of linear systems of second-order ordinary differential equations