Abstract: To address the demand for remote and minimally invasive neurosurgeries for pain relief in high-intensity confrontation on the battlefield, an embodied intelligence platform for an AI-enhanced OctoMag system is proposed based on vascular-targeted drug delivery interventions using magnetic micro-nanorobots. By integrating a “magneto-electric” model, the deep reinforcement learning algorithm PPO (Proximal Policy Optimization), and domain randomization training strategies, the platform enhances the autonomous learning and adaptability of the OctoMag system in interacting with its environment. The simulation results demonstrate that the platform maintains high-precision control performance within the range of human arterial blood viscosity, with a normalized mean positioning error not exceeding 0.06. Under extreme viscosity conditions (0.007 Pa·s), the control accuracy is improved by 88.4% compared with that of the PPO algorithm without domain randomization.