Fast Online Adaptive Neural MPC via Meta-Learning

2025-04-23

AI Summary

To address the slow online adaptation and high computational overhead of data-driven neural model predictive control (MPC) under model uncertainty, this paper proposes a meta-learning-based online adaptive neural MPC framework. The method integrates Model-Agnostic Meta-Learning (MAML) into neural MPC for the first time, enabling rapid residual dynamics correction using only a few online samples and one- or few-step gradient updates. By decoupling residual dynamics modeling from nonlinear optimization, the framework balances modeling accuracy with millisecond-level real-time performance. Leveraging the L4CasADi efficient solver, it demonstrates superior performance over baseline methods in Van der Pol, Cart-Pole, and 2D quadrotor simulations: residual model adaptation speed improves by 3–5×, prediction error decreases by over 40%, and closed-loop robotic control requirements for both real-time operation and robustness are fully satisfied.

Technology Category

• Machine Learning: Adversarial Learning & Robustness
• Multiagent Systems: Multiagent Learning
• Intelligent Robots: State Estimation

Application Category

• User Modeling, Personalization and Recommendation: On-Device user modeling, personalization, and recommendation
• Systems and Infrastructure for Web, Mobile and WoT: Applied ML and AI for Web-based mobile applications
• Responsible Web: Machine-in-the-loop, human agency and autonomy

Abstract

Data-driven model predictive control (MPC) has demonstrated significant potential for improving robot control performance in the presence of model uncertainties. However, existing approaches often require extensive offline data collection and computationally intensive training, limiting their ability to adapt online. To address these challenges, this paper presents a fast online adaptive MPC framework that leverages neural networks integrated with Model-Agnostic Meta-Learning (MAML). Our approach focuses on few-shot adaptation of residual dynamics - capturing the discrepancy between nominal and true system behavior - using minimal online data and gradient steps. By embedding these meta-learned residual models into a computationally efficient L4CasADi-based MPC pipeline, the proposed method enables rapid model correction, enhances predictive accuracy, and improves real-time control performance. We validate the framework through simulation studies on a Van der Pol oscillator, a Cart-Pole system, and a 2D quadrotor. Results show significant gains in adaptation speed and prediction accuracy over both nominal MPC and nominal MPC augmented with a freshly initialized neural network, underscoring the effectiveness of our approach for real-time adaptive robot control.

Problem

• Adapting MPC online with minimal data and computation
• Improving robot control under model uncertainties
• Enhancing real-time prediction and control accuracy

Innovation

• Meta-learning for few-shot adaptation of dynamics
• Neural networks integrated with MAML framework
• L4CasADi-based MPC for real-time control