An inter-row navigation system design for a safflower picking robot

Accurate inter-row navigation for safflower harvesting robots remains challenging in curved crop rows due to nonlinear terrain effects, sensor uncertainties, and limited interaction awareness between robots and crops. To address these issues, this study proposes a multi-sensor fusion navigation framework integrating GNSS, IMU, wheel encoders, and a nearside divider pressure sensor. A pure pursuit-based kinematic model is combined with an adaptive fuzzy PID controller to achieve global trajectory tracking, while a pressure-based local correction mechanism is introduced to incorporate crop–robot interaction feedback for local path adjustment under curved-row conditions. Field experiments demonstrate that the proposed system achieves a steady-state tracking error below 40 mm at 1 km/h, with steady-state errors ranging from 33 to 42 mm across different initial conditions. In the tested curved-row scenario, the proposed method maintains an average tracking error below 70 mm, indicating that pressure-based local feedback can contribute to trajectory correction under curved-row conditions. Comparative evaluations against PID, pure pursuit, and fuzzy PID controllers confirm that the proposed framework significantly improves tracking accuracy and adaptability in nonlinear agricultural environments. Ablation results further verify that the pressure-based interaction module is a critical contributor to performance improvement in curved-row navigation. The results demonstrate that integrating global navigation with local physical interaction feedback provides a robust and practical solution for autonomous inter-row agricultural operation in complex field environments.