RL-trained microrobots navigate simulated capillaries to unblock blood flow

A team led by Jannik Drotleff at the University of Stuttgart developed a physically grounded simulation of a blood capillary network, incorporating realistic hydrodynamic flow fields, explicit red blood cell dynamics, and anatomically derived branching geometry. Using deep reinforcement learning, they trained agents to navigate the complex environment via chemotaxis. The study systematically mapped the physical limits of navigation across robot size and swimming speed, revealing a forbidden regime where Brownian motion and flow overcome propulsion. Successful agents independently discovered multiple universal strategy types, including run-and-rotate and energy-efficient search-and-sit policies, regardless of robot parameters.

Most notably, without any retraining, these agents performed targeted blocking and unblocking of capillary flow, restoring throughput to healthy baseline levels. This demonstrates that RL agents can generalize learned policies to intervention tasks in complex biological environments. The results establish reinforcement learning as a viable framework for developing autonomous microrobotic intervention strategies, with potential applications in targeted drug delivery and thrombolysis. The paper is available on arXiv (2606.26154).