Physics-Informed 3D Simulation and Data Architecture for AI-Driven Magnetic Micro-robots in Endovascular Navigation

Abstract

Achieving precise navigation within the human circulatory system requires a robust understanding of the complex micro-scale forces at play. This paper presents a comprehensive theoretical and computational framework for the navigation of a composite magnetic micro-robot, composed of a ferromagnetic-polymer matrix, within a 3D vascular environment. The mathematical model integrates magnetic gradient actuation, non-linear hydrodynamic drag, apparent weight, and short-range forces, including electrostatic interactions and Hertzian contact mechanics. Crucially, to bridge the gap between biomechanical simulation and Artificial Intelligence, the physics-informed MATLAB environment was equipped with a high-frequency data extraction architecture. This system dynamically records teleoperated navigation as Markov Decision Process (MDP) tuples (St, At, Rt), using a dense reward function that heavily penalizes simulated tissue collisions. Results demonstrate not only the feasibility of navigating against dynamic hemodynamic conditions using multi-axial magnetic gradients, but also the successful generation of high-fidelity datasets. This provides a robust, pre-computed foundation for training autonomous medical interventions via Offline Reinforcement Learning, paving the way for safer and more accessible endovascular therapies.