Learning control of micro-robots in vascular systems

Abstract Autonomous navigation of micro-robots through the human vascular system is a challenging control problem due to complex haemodynamics, non-Newtonian blood flow, and highly dynamic obstacles. This paper proposes an Adaptive Turbulence-Aware Reinforcement Learning (ATARL) framework for micro-robot navigation in simulated vascular environments. The framework integrates a turbulence risk estimator with an adaptive exploration strategy, which allows the micro-robot to balance path efficiency and collision avoidance under varied flow conditions. A Unity-based simulation platform that replicates pulsatile non-Newtonian blood flow is developed to train and evaluate the proposed controller. Experimental results demonstrate that the ATARL-controlled micro-robot achieves a navigation success rate of over 90% across simple, branching, and turbulent flow scenarios. These results highlight the potential of the proposed framework for medical applications such as targeted drug delivery, plaque removal, and vascular diagnostics.