基於ROS系統之輪椅舞導入生成式AI路徑

Abstract

本研究旨在因應高齡化社會對輔具日益增長的需求，以及身心障礙者對自我實現的渴望，提出一套創新的電動輪椅舞解決方案。有鑑於傳統手推輪椅舞對於舞者體力要求高，且可能導致上肢勞損與受傷風險。 本論文的核心貢獻在於開發一個基於生成式人工智慧的智能電動輪椅舞編舞系統。該系統透過音樂分析、參數計算、動作生成、路徑生成及最終輸出等五個階段進行舞步規劃。其中，音樂分析模組利用Spotify API和ChatGPT API辨識歌曲的BPM、風格與情感特徵。動作生成模組基於這些音樂特徵，編排預設的輪椅基本動作，以確保舞步的多樣性與流暢性。路徑生成模組則將舞步轉換為具體空間座標路徑，並計算出速度以及角速度傳送給輪椅執行。 為實現輪椅舞步控制，本系統以ROS (Robot Operating System) 作為整合平台。電動輪椅硬體採用實驗室自行研發改裝之雙動力輪電動輪椅，其上控制器由Intel NUC迷你電腦與Arduino MEGA 2560組成，下控制器則包含DSP，負責動力輪馬達的精確控制。ROS與Arduino之間則透過rosserial套件進行資料傳輸，將上位機生成的舞步路徑轉換為輪椅可執行的速度指令。 本研究的成果預期將降低輪椅舞的參與門檻，促進身心障礙者的自我表達與社會融入，並為輪椅在藝術與復健領域的跨域應用開拓新的可能性。

This study aims to address the growing demand for assistive devices in an aging society and the desire for self-fulfillment among individuals with physical disabilities by proposing an innovative solution for electric wheelchair dance. Traditional manual wheelchair dance requires substantial physical effort from dancers and may result in upper limb fatigue and injury risks. The core contribution of this thesis is the development of an intelligent electric wheelchair dance choreography system based on generative artificial intelligence. The system plans dance steps through five stages: music analysis, parameter calculation, movement generation, path generation, and final output. The music analysis module utilizes the Spotify API and ChatGPT API to accurately identify a song’s BPM, style, and emotional characteristics. The movement generation module intelligently arranges preset wheelchair basic movements according to these music features to ensure diversity and fluency of dance steps. The path generation module then translates the dance steps into specific spatial coordinates, employing cubic spline interpolation for smoothing. For safety and obstacle avoidance, the system integrates collision detection based on binarized maps and local path adjustments. To realize wheelchair dance control, the system uses ROS (Robot Operating System) as the integration platform. The hardware consists of a laboratory-developed dual-drive electric wheelchair. The upper-level controller comprises an Intel NUC mini computer and an Arduino MEGA 2560, while the lower-level controller includes a DSP responsible for precise control of the drive wheel motors. ROS communicates with Arduino via the rosserial package, transmitting the generated dance path from the host computer and converting it into executable speed commands for the wheelchair. The outcomes of this research are expected to significantly lower the barriers to participating in wheelchair dance, promote self-expression and social integration for individuals with physical disabilities, and open new possibilities for the interdisciplinary application of wheelchairs in the fields of art and rehabilitation.