| 摘要: | 本研究針對一款六足蜘蛛型機器人進行結構改造，增設可控電磁吸附腳部模組，使其能在水平、傾斜與垂直鋼構表面穩定附著並執行攀爬；為提升通行可靠度，提出以視覺避障為核心的足點規劃與步態控制：採用 YOLOv4-Tiny 偵測 RGB 影像中的螺栓，與深度影像融合計算螺栓三維座標，依橫向偏差驅動「偵測—側向對齊—前進」之閉環流程，主動將落足點置於相鄰螺栓間隙以避免踩踏與吸附干擾。在步態設計上，本研究著重於改良三角步態以適應傾斜與垂直場域，並針對電磁鐵釋放、著地時序與分組順序進行規劃；同時導入支撐多邊形理論作為穩定性分析基礎，結合IMU 量測，量化評估俯仰、橫滾姿態變化與行走偏差。實驗於 45° 與 90° 鋼板環境進行行走穩定性、釋放順序、姿態偏移、路徑偏差、移動效率與越障等測試，結果顯示所提視覺避障與改良三角步態能有效避開螺栓並維持姿態穩定，驗證六足磁吸平台於鋼構表面之可行性與可靠度。綜上所述，本研究完成一套整合電磁吸附、視覺避障與步態控制之六足攀爬機器人控制方法，為後續在橋梁、儲槽等大型鋼構的自動化巡檢奠定基礎。;This study redesigns a hexapod spider-type robot by integrating controllable electro-magnetic adhesion modules at the feet, enabling stable attachment and locomotion on horizontal, inclined, and vertical steel surfaces. To improve traversal reliability, we propose a vision-centric foothold planning and gait control framework: bolts are detected in RGB images using YOLOv4-Tiny, fused with depth measurements to recover 3-D bolt coordinates, and a detect–align–advance closed loop is triggered based on lateral offset so that footholds are deliberately placed in inter-bolt gaps, preventing foot–bolt contact and adhesion interference. On the locomotion side, we adapt the tripod gait for inclined and vertical operation and schedule groupwise electromagnet release/landing sequences. Stability analysis is grounded in the support-polygon theory and quantified via servo feedback and IMU measurements (pitch/roll and path deviation). Experiments on 45° and 90° steel plates evaluate walking stability, release sequencing, attitude drift, path deviation, locomotion efficiency, and obstacle crossing. Results show that the proposed visual obstacle avoidance, coupled with the modified tripod gait, effectively avoids bolts and maintains stable posture, demonstrating the feasibility and reliability of magnetically adhered hexapod locomotion on steel structures. Overall, this work delivers an integrated methodology that combines magnetic adhesion, vision-based obstacle avoidance, and gait control, laying the groundwork for automated inspection of large steel assets (e.g., bridges and storage tanks) with future extensions to autonomous navigation and defect recognition. |
