3D Closed‐Loop Navigation of Magnetic Microswimmers Using Vision‐Derived Depth Feedback

Despite growing interest in autonomous microswimmer navigation, achieving closed‐loop 3D control remains challenging because simple, real‐time depth feedback compatible with standard microscopy is unavailable. This work presents a unified multimodal control framework integrating Model Predictive Control (MPC) with A* path planning for both surface‐bound and free‐swimming locomotion of planar magnetic microswimmers. A propulsion‐only depth regulation strategy uses a signed Fourier‐domain blur metric from a single monocular microscope image to estimate axial position without magnetic gradients or volumetric imaging. This vision‐based method enables real‐time depth estimation and waypoint tracking with minimal computational cost. In experiments, the proposed controller achieved 3D closed‐loop navigation with a median planar error of 12–15 μm and overshoot of 20 ± 5 μm (mean ± SD, n ≥ 20). During fixed‐focus 3D swimming, depth regulation was maintained within an axial working window of approximately 200 μm. For surface locomotion, A*‐guided MPC completed obstacle‐rich routes about 40% faster than reactive MPC while maintaining 100% target‐reach success when a collision‐free path existed. These results demonstrate robust multimode transitions, depth‐aware 3D control, and high‐precision navigation using only propulsion and monocular vision, providing a foundation for future in vitro and ophthalmic applications.