智慧型跟隨式醫療機器人

Abstract

中文摘要 在現今醫療環境中，隨著高齡化社會來臨，護理人員短缺與工作負荷過重已成為全球性的醫療危機。護理師在日常工作中，需頻繁推動裝載藥品、醫療器材或病歷的護理推車穿梭於病房與護理站之間，這不僅消耗大量體力，更佔用雙手，降低了照護效率。為解決此問題，本計畫將開發一台「智慧型跟隨式醫療機器人」。

本系統將傳統被動式的護理推車與自主移動機器人（AMR）技術結合，如圖一所示。計劃採用分層控制設計：底盤採用兩輪差速驅動搭配萬向輪的設計，具備原地轉向與靈活移動的特性，以適應狹窄的醫院走廊，裝備 NVIDIA Jetson Orin AGX 作為核心運算單元，搭載 ROS 2（Robot Operating System 2）機器人作業系統，整合 LiDAR 光達與深度相機進行感測器融合（Sensor Fusion），實現複雜醫院環境下的 SLAM 建圖、路徑規劃、動態避障及精準人員跟隨功能。上層架構方面，本計畫創新導入大型語言模型（LLM） 作為上位機的核心大腦，取代傳統僵化的指令識別。透過搭載於車身的麥克風陣列與視覺感測器，醫療車不僅能透過電腦視覺技術（Computer Vision）鎖定並自動跟隨特定護理人員，更能理解護理師的自然語言指令（如：「請跟著我到 305 病房」、「請停在這裡等我」），並執行對應的導航或待命動作。

針對醫療推車高負載的需求，本研究依據實際醫療場域數據（車體含設備滿載約 120 公斤），進行了嚴謹的動力學分析與馬達選型，選用具備高解析度編碼器之伺服驅動輪（ZLLG65ASM250-L V3.0），確保機器人能在 1.65 m/s 的人類快步走速度下平穩運行，並具備足夠的扭力克服醫院內的斜坡與門檻。

在人機互動與智慧化應用層面，本系統於上位機導入大型語言模型技術。該模組具備四大核心功能：(1) 自然語言導航控制：護理人員可透過語音指令直接控制底盤動作；(2) 醫病溝通輔助：作為查房助手與病患進行基礎衛教問答；(3) 醫療系統整合：串接醫院資訊系統（HIS）以即時獲取並朗讀病歷摘要；(4) 物聯網監測：連接外部生理量測儀器，即時分析病患數據。本研究透過軟硬體的高度整合，期望打造出具備「眼（視覺導航）」、「腦（LLM 認知）」與「腳（底盤）」的智慧醫療助理，為未來的智慧醫院（Smart Hospital）提供具體可行的自動化解決方案。 Abstract With the intensifying global aging population, healthcare systems are facing unprecedented pressure. Shortages of nursing staff and excessive workloads have become common challenges in medical facilities worldwide. In daily nursing routines, nurses are required not only to move frequently between wards and nursing stations but also to transport heavy medical carts loaded with medicines, consumables, and electronic devices. This not only reduces work efficiency but also causes long-term occupational musculoskeletal injuries to nursing personnel.

This study aims to develop a "Smart Medical Follower Robot" by integrating autonomous navigation technology with Large Language Models (LLMs) to alleviate the physical burden on medical staff and improve the efficiency of medical information processing. In terms of system architecture, this study adopts a hierarchical control design. The chassis motion control layer utilizes the NVIDIA Jetson Orin AGX as the core computing unit, running on the Robot Operating System 2 (ROS 2). It integrates LiDAR and depth cameras for sensor fusion, achieving Simultaneous Localization and Mapping (SLAM), path planning, dynamic obstacle avoidance, and precise human following in complex hospital environments.

Addressing the high-load requirements of medical carts, this study conducted rigorous dynamic analysis and motor selection based on actual medical field data (a fully loaded cart weighs approximately 120 kg). We selected ZLLG65ASM250-L V3.0 servo drive wheels equipped with high-resolution encoders. This ensures the robot can operate smoothly at a human brisk walking speed of 1.65 m/s and possesses sufficient torque to overcome ramps and thresholds within the hospital.

Regarding human-robot interaction and intelligent applications, the system incorporates Large Language Model technology in the upper-level host. This module features four core functions: (1) Natural Language Navigation Control: Nurses can control chassis movements directly via voice commands; (2) Doctor-Patient Communication Assistance: Acting as a ward round assistant to conduct basic health education Q&A with patients; (3) Medical System Integration: Connecting with the Hospital Information System (HIS) to retrieve and read out medical record summaries in real-time; and (4) IoT Monitoring: Connecting to external physiological measurement instruments to analyze patient data instantly. Through high-level integration of software and hardware, this study expects to create a smart medical assistant equipped with "Eyes (Visual Navigation)","Brain (LLM Cognition)" and "Legs (Robust Chassis)",providing a concrete and feasible automation solution for future Smart Hospitals.