Autonomous Visual Navigation for Quadruped Robot in Farm Operation

Abstract

In agricultural robotics, the development of practical and scalable autonomy solutions is crucial for operational efficiency and effectiveness. Autonomous navigation, the cornerstone capability for robots, is the basis for any agricultural tasks such as crop monitoring, weed management, and crop transportation. This study centers on developing foundation navigation autonomy for quadruped robots in agricultural settings with experimental validations in strawberry field scenarios. Our approach removes the reliance on costly sensors, opting instead for a vision-based system that ensures robustness and reliability in navigating strawberry fields. Through field testing, we demonstrate that our navigation autonomy achieves precise furrow entry, tracking, and exit. The robot exhibits exceptional mobility, adeptly handling muddy and uneven terrain with the aid of vision. Furthermore, our foundational autonomy framework effectively detects obstacle dimensions for both planning and control to ensure safe interactions between robots and human workers in the field. This work not only presents a leap in agricultural robotics autonomy but also lays the groundwork for broader applications of quadruped robots in complex, real-world scenarios.