本研究提出一種基於微型機器學習可進行AI智能自主移動避障的機器人GLEWBOT-VISION。該機器人解決了在遇到磁磚較大缺損處時吸附失效的問題。GLEWBOT-VISION系統利用FOMO視覺辨識模型應用於微型鏡頭中，能在偵測到磁磚缺陷後立即執行相應的避障動作。系統採用了輕量化的FOMO模型，能夠即時偵測出不同位置和形狀的瓷磚缺陷。 為了驗證系統的有效性，本研究進行了多組實驗，包括視覺辨識模組測試和實際應用情況下的測試。實驗結果顯示，整體模型的精準度高達95%，在真實牆面上應用於不同方向偵測時的精準度也達到95%。在不同的鏡頭覆蓋範圍下，其精準度分別為：僅覆蓋25%面積的缺陷時為80%、覆蓋一半面積的缺陷時為90%、完全覆蓋時為95%。後續實驗中展示了系統在牆面上應用的實際效果。 本研究的創新之處在於將微型機器學習、FOMO視覺辨識與音訊分析技術相結合，實現了GLEWBOT-VISION的自主移動避障功能。這不僅降低了GLEWBOT-VISION吸附於磁磚缺陷處時吸附失效的可能性，還減少了對專業技術人員的依賴，降低了人力成本。同時，系統設計考慮了現場應用的便利性和靈活性，能夠檢測到不同位置和形狀的目標物。未來，該系統還可以進一步擴展應用於其他類型的智能移動設備中，具有廣泛的發展潛力和應用價值。 ;This study presents GLEWBOT-VISION, an AI-driven autonomous obstacle avoidance robot system integrating the GLEWBOT with the FOMO visual recognition model and miniature machine learning technology. It addresses GLEWBOT′s suction failure on defective tiles by using the FOMO model with a miniature camera for real-time defect detection and obstacle avoidance. The lightweight FOMO model effectively detects various tile defect positions and shapes. Multiple experiments validated the system′s effectiveness, showing a 95% accuracy overall, consistent in real-world applications. Accuracy varied with camera coverage: 80% at 25% coverage, 90% at half, and 95% at full defect coverage. Subsequent tests confirmed the system′s real-world performance on wall surfaces. This study innovatively combines miniature machine learning, FOMO visual recognition, and audio analysis, enabling autonomous obstacle avoidance for GLEWBOT-VISION. It reduces suction failures, lowers reliance on technicians, and cuts labor costs. The design also ensures convenience and flexibility for field applications. This system has potential for further development and application in other intelligent mobile devices.
