Reinforcement Learning-Controlled Magnetic Colloid for Micro-cargo Transport

Precise, contactless transport of microscale cargo to arbitrary locations is a critical challenge in microfluidics, nanomedicine, and microassembly. While microrobots often rely on direct grasping or cargo functionalization, we present a contactless strategy that exploits programmable hydrodynamic flow. A magnetically translated and rotated colloidal microrotor generates tailored flow fields in its vicinity, enabling the contactless transport of similarly sized cargo particles in a fluid environment. Based on this hydrodynamic transport mechanism, we develop a feedback control system that combines real-time visual tracking with a reinforcement learning framework. The system adaptively regulates the direction of rotation of a rotating magnetic field to dynamically optimize the microrotor's motion and the induced hydrodynamic forces, enabling targeted delivery of microscale cargo without physical contact or chemical modification. Further progressive training with path planning algorithms allows the system to achieve complex tasks such as obstacle avoidance and maze solving. Our method establishes a versatile platform for noncontact microscale manipulation in fluid environments.