A Mobile Magnetic Manipulation Platform for Gastrointestinal Navigation with Deep Reinforcement Learning Control

TL;DR

The paper tackles the challenge of precise magnetic navigation for targeted gastrointestinal drug delivery by replacing model-heavy calibration with a learning-based controller. It introduces a portable mobile platform consisting of a four-electromagnet end-effector mounted on a UR5 robot, paired with a Soft Actor-Critic (SAC) controller trained in a physics-based simulator and refined via a sim-to-real pipeline to operate without explicit field-model calibration. The approach achieves millimeter-scale tracking accuracy in 2D and demonstrates performance over a large, clinically relevant workspace (approximately $30\ \text{cm} \times 20\ \text{cm}$) with a 7 mm magnetic capsule, outperforming fixed-current and PID baselines. The proposed framework reduces setup time to roughly $45$ minutes, enabling rapid deployment in GI navigation scenarios, though limitations include 2D validation and reliance on optical guidance; future work aims to extend to 3D navigation and integrate clinically relevant imaging modalities.

Abstract

Targeted drug delivery in the gastrointestinal (GI) tract using magnetic robots offers a promising alternative to systemic treatments. However, controlling these robots is a major challenge. Stationary magnetic systems have a limited workspace, while mobile systems (e.g., coils on a robotic arm) suffer from a "model-calibration bottleneck", requiring complex, pre-calibrated physical models that are time-consuming to create and computationally expensive. This paper presents a compact, low-cost mobile magnetic manipulation platform that overcomes this limitation using Deep Reinforcement Learning (DRL). Our system features a compact four-electromagnet array mounted on a UR5 collaborative robot. A Soft Actor-Critic (SAC)-based control strategy is trained through a sim-to-real pipeline, enabling effective policy deployment within 15 minutes and significantly reducing setup time. We validated the platform by controlling a 7-mm magnetic capsule along 2D trajectories. Our DRL-based controller achieved a root-mean-square error (RMSE) of 1.18~mm for a square path and 1.50~mm for a circular path. We also demonstrated successful tracking over a clinically relevant, 30 cm * 20 cm workspace. This work demonstrates a rapidly deployable, model-free control framework capable of precise magnetic manipulation in a large workspace,validated using a 2D GI phantom.

Figures (10)

• Figure 1: Overview of our mobile magnetic manipulation platform and control system.
• Figure 2: (a) Electromagnetic end-effector including the four-electromagnet array, 3D-printed mount and a RealSense camera. (b) Front view and bottom view of the magnetic capsule.
• Figure 3: Schematics of the electromagnet array control.
• Figure 4: Overview of our sim-to-real pipeline. A policy is pre-trained in a physics-based simulation and then few-shot fine-tuned on our mobile magnetic manipulation platform, yielding a fine-tuned policy for real-time control on the physical system.
• Figure 6: Continuous trajectory spanning a 30 cm × 20 cm area, the purple line is the desired trajectory.