Automating Robot Motion Planning for Magnetic Resonance Navigation Using Q-Learning

Abstract

Magnetic resonance navigation (MRN) is promising to drive Micro/Nano robots in blood vessels for some potential clinical applications. It requires real-time processing under constraints of interleaved imaging and steering MR sequence. Efficient and fast motion planning is desired for Micro/Nano robots navigated during MRN. In this paper, a novel navigation method is proposed for guiding tiny robots in blood vessels. An accurate segmentation of blood vessels on MRA images with Bayesian inference is used for extracting exact vessels at first. Then, a reinforcement learning mechanism is designed to navigate robot agents within vascular network. Experimental results demonstrate that the proposed method can simulate the robot navigation on MRA image vessels. The path between starting point and ending point can be automatically identified.