A team at The Chinese University of Hong Kong has demonstrated AI tracking that can locate millimeter-scale wireless magnetic robots while a magnetic field drives them. The researchers report millimeter-level localization at tissue depths up to 10 centimeters and closed-loop navigation in a goat model without magnetic shielding.
The robots are difficult to locate because the magnetic fields that move them overwhelm the much weaker signal used to track them. The problem grows as a robot shrinks: the paper says magnetic-signal tracking had been limited to centimeter-scale devices.
The system separates the interference problem in two stages. A sensor array picks up the robot’s weak magnetic signal along with interference from the stronger field moving it. A transformer model estimates that interference and filters it from the sensor readings; a second model uses a short sequence of the cleaned signals to estimate the robot’s three-dimensional position. Those estimates feed back to the controller so it can guide the robot while it moves.
In the goat model, CUHK says the team navigated the robots through the space around the spinal cord (the spinal subarachnoid space) and the bile duct, without magnetic shielding; estimated trajectories closely matched X-ray observations.
The goat work is a preclinical navigation demonstration, not evidence of treatment or human use. A human study would be needed before the system’s performance in clinical procedures could be judged.
Sources
- Hu et al., “AI-driven actuation-compatible tracking for closed-loop navigation of miniature robots in vivo,” Science Robotics (2026)
- The Chinese University of Hong Kong, research announcement (24 September 2026)
- Illustrative diagram, not from the featured study: Hakan Ceylan, Immihan C. Yasa, Ugur Kilic, Wenqi Hu and Metin Sitti, “Essentials for a microrobot to function in vivo”, CC BY 3.0; unchanged.
