Trajectory Tracking and Stabilization of Quadrotors Using Deep Koopman Model Predictive Control

2025-08-19

AI Summary

This work addresses the challenge of high-precision trajectory tracking and point stabilization for quadrotors under strongly nonlinear dynamics. We propose a real-time control framework that synergistically integrates deep Koopman operator theory with model predictive control (MPC). Leveraging data-driven learning, our method identifies an invertible mapping from the original nonlinear state space to a linear latent space, thereby transforming the control problem into a computationally efficient linear MPC optimization while preserving modeling fidelity. Key contributions include: (i) the first integration of a deep Koopman operator into a closed-loop flight control architecture, enabling end-to-end differentiable latent-space modeling and online control; and (ii) elimination of the prohibitive computational burden associated with conventional nonlinear MPC. Simulation results demonstrate a 32% reduction in tracking error and a 78% decrease in per-step optimization time compared to baseline nonlinear MPC, confirming the method’s feasibility and superiority on resource-constrained embedded flight control platforms.

Technology Category

• Search and Optimization: Learning to Search
• Intelligent Robots: State Estimation
• Computer Vision: Learning & Optimization for CV

Application Category

• Responsible Web: Machine-in-the-loop, human agency and autonomy
• Economics, Online Markets and Human Computation: LLM based quality controls for crowd work
• Systems and Infrastructure for Web, Mobile and WoT: Applied ML and AI for Web-based mobile applications

Abstract

This paper presents a data-driven control framework for quadrotor systems that integrates a deep Koopman operator with model predictive control (DK-MPC). The deep Koopman operator is trained on sampled flight data to construct a high-dimensional latent representation in which the nonlinear quadrotor dynamics are approximated by linear models. This linearization enables the application of MPC to efficiently optimize control actions over a finite prediction horizon, ensuring accurate trajectory tracking and stabilization. The proposed DK-MPC approach is validated through a series of trajectory-following and point-stabilization numerical experiments, where it demonstrates superior tracking accuracy and significantly lower computation time compared to conventional nonlinear MPC. These results highlight the potential of Koopman-based learning methods to handle complex quadrotor dynamics while meeting the real-time requirements of embedded flight control. Future work will focus on extending the framework to more agile flight scenarios and improving robustness against external disturbances.

Problem

• Accurate trajectory tracking for quadrotor systems
• Stabilization of nonlinear quadrotor flight dynamics
• Real-time control optimization with reduced computation

Innovation

• Deep Koopman operator linearizes quadrotor dynamics
• Model predictive control optimizes finite horizon actions
• Data-driven framework enables real-time embedded implementation