Development of Wilted Leaf Pose Detection System by Using Axiomatic Design

This study presents a wilted leaf pose detection system for agricultural robots, developed based on the axiomatic design methodology. Conventional cutting-point detection systems rely on deep learning and geometric algorithms with strong interdependencies, lacking system-level uncertainty control. To address this, function requirements and design parameters were hierarchically mapped into a four-level design matrix, defining ten leaf modules to ensure independence. An information flow control (IFC) mechanism was introduced at key coupled modules to manage uncertainty propagation, and a qualitative information content (QIC) measure was proposed to heuristically interpret information content based on prior assumptions and testability. Experiments demonstrated that the proposed IFC reduced mean position and rotation errors by 9.7% and 12%, respectively. Compared to the ICP (Iterative Closet Point)-based baseline, our method achieved a 4.4-fold increase in sucess rate under occluded conditions, demonstrating that the QIC measure represents information content.