Zero-Shot Function Encoder-Based Differentiable Predictive Control

TL;DR

This paper presents a novel differentiable framework combining a function encoder-based neural ODE with predictive control for zero-shot adaptive control of nonlinear systems, enabling fast adaptation without retraining.

Contribution

It introduces FE-NODE and DPC to model nonlinear dynamics and learn control policies across parameters, eliminating the need for online optimization.

Findings

Demonstrates efficiency and accuracy across various nonlinear systems.

Enables zero-shot adaptation to new systems without retraining.

Reduces computational cost compared to classical MPC.

Abstract

We introduce a differentiable framework for zero-shot adaptive control over parametric families of nonlinear dynamical systems. Our approach integrates a function encoder-based neural ODE (FE-NODE) for modeling system dynamics with a differentiable predictive control (DPC) for offline self-supervised learning of explicit control policies. The FE-NODE captures nonlinear behaviors in state transitions and enables zero-shot adaptation to new systems without retraining, while the DPC efficiently learns control policies across system parameterizations, thus eliminating costly online optimization common in classical model predictive control. We demonstrate the efficiency, accuracy, and online adaptability of the proposed method across a range of nonlinear systems with varying parametric scenarios, highlighting its potential as a general-purpose tool for fast zero-shot adaptive control.