Zero-Shot Function Encoder-Based Differentiable Predictive Control

2025-11-07

AI Summary

Adaptive control of parametric nonlinear dynamical systems typically requires online optimization or model retraining, hindering real-time deployment and generalization across system parameters. Method: This paper proposes a zero-shot adaptive control framework that integrates a Function Encoder into Neural Ordinary Differential Equations (FE-NODE) to implicitly encode system parameters, coupled with Differentiable Predictive Control (DPC) for end-to-end, offline, self-supervised policy learning—eliminating online optimization entirely. Contribution/Results: (1) The function encoder enables plug-and-play generalization across unseen parameter configurations; (2) differentiable closed-loop optimization replaces computationally expensive online Model Predictive Control (MPC) solvers. Evaluated on multiple canonical nonlinear systems, the approach achieves significant improvements in control deployment speed, cross-parameter generalization accuracy, and real-time adaptability. It establishes a new paradigm for efficient, sample-efficient adaptive control of parametric dynamical systems.

Technology Category

• Search and Optimization: Learning to Search
• Intelligent Robots: Learning & Optimization for ROB
• Cognitive Modeling & Cognitive Systems: Adaptive Behavior

Application Category

• Economics, Online Markets and Human Computation: Cost models of using LLMs in production systems
• Systems and Infrastructure for Web, Mobile and WoT: Experiences and lessons learnt from Web-based algorithms and system deployments
• User Modeling, Personalization and Recommendation: On-Device user modeling, personalization, and recommendation

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.

Problem

• Develops zero-shot adaptive control for nonlinear dynamical systems
• Eliminates online optimization in model predictive control
• Enables control policy adaptation without system retraining

Innovation

• Function encoder neural ODE models system dynamics
• Differentiable predictive control learns offline policies
• Zero-shot adaptation eliminates retraining and online optimization