輪腳複合機器人上具視覺回授與多輪腳觸地點之爬樓梯策略

Abstract

輪腳複合機器人兼具輪式高速移動與足式地形適應的優點，但由於其特殊的腳部構型，會面臨模式切換複雜、觸地點不易判定和無法快速靈活揮腳等諸多挑戰。而在爬樓梯時因高低落差大與空間受限，更需要確保機器人的穩定性並避免與地形碰撞。為解決上述問題，本研究針對實驗室所開發之第三代輪腳複合機器人，提出一套結合視覺資訊與考慮多輪腳觸地點的爬樓梯策略，以提升機器人在非連續地形中垂直移動的穩定性與自主性。 由於輪腳機構的複雜性與特殊性，本研究首先實現輪腳模組的逆向運動學並提出可變觸地點的計算方法，建立考慮輪框滾動的運動學規劃方式。接著，從工作空間分析、穩定性調整、踏點選擇與揮腳軌跡等面向，建立系統化的爬樓梯策略規劃流程。最後，利用深度相機獲取環境之點雲資訊，進行平面位置估測得到樓梯尺寸，以實現即時軌跡規劃的需求。 在實驗部分，於室內外多種不同尺寸的樓梯上進行測試，驗證所提出之策略的可行性與適應性，並分析過程中的控制與估測誤差，探討策略成功與失敗時的原因，以及比較室內外不同環境對策略表現造成的差異與影響。結果顯示，雖然於室外環境的估測誤差較室內環境大，但在能維持一定估測精度的情況下，所提出之爬樓梯策略於室內與室外樓梯環境皆能穩定完成爬升。 綜上所述，本研究驗證了在輪腳複合機器人上具視覺回授的爬樓梯策略的應用潛力，為此類機器人的運動行為開發以及未來多地形自主導航奠定良好基礎。

Leg-wheel robots combine the advantages of high-speed mobility in wheeled mode and adaptability to complex terrains in legged mode. However, their unique leg structures pose several challenges, including complex mode transitions, difficulty in calculating contact points, and limited ability for fast and flexible leg swings. When climbing stairs, the large height differences and spatial constraints further demand enhanced stability and collision avoidance. To address these challenges, this study proposes a stair-climbing strategy for the third-generation leg-wheel robot developed in our laboratory, integrating visual information and considering varying contact points to improve the robot's stability and autonomy during vertical movements on discontinuous terrains. Given the complexity and special characteristics of the leg-wheel mechanism, this research first implements the inverse kinematics of the leg-wheel module and proposes a computation method for varying contact points, establishing a kinematic planning approach that accounts for wheel rolling. Subsequently, a systematic stair-climbing strategy planning framework is developed, covering workspace analysis, stability adjustment, foothold planning, and swing trajectory design. Finally, a depth camera is utilized to acquire point cloud data of the environment, enabling planar position estimation for stair dimension measurement to support real-time trajectory planning. For experimental validation, tests were conducted on indoor and outdoor stairs of various sizes to verify the feasibility and adaptability of the proposed strategy. Control and estimation errors during the climbing process were analyzed to investigate the causes of success and failure, and differences in strategy performance between indoor and outdoor environments were discussed. The results indicate that, although estimation errors are larger in outdoor environments compared to indoor ones, the proposed strategy can still achieve stable stair climbing in both conditions when a certain extent of estimation accuracy is maintained. In summary, this study demonstrates the potential of a stair-climbing strategy using visual feedback for the leg-wheel robot, providing a solid foundation for the development of locomotion behaviors and future autonomous navigation across diverse terrains.