Autonomous Navigation Algorithm for a Rocker-Bogie Vehicle Using Stereo Vision-Based SLAM in Hass Avocado Agricultural Fields

Jaramillo Ocampo N.M.; Salazar M.B.S.; Felipe J.L.B.; Alvarez N.M.R.; Rivas J.R.R.; Vilcarromero J.J.F. · 2025 8th International Conference on Robotics, Control and Automation Engineering, RCAE 2025 (2025) · DOI: 10.1109/rcae66389.2025.11355282

Resumen

This paper presents an autonomous navigation system for a non-holonomic six-wheeled agricultural rover equipped with a rocker-bogie suspension, specifically designed to operate in semi-structured crop field environments. The system leverages onboard stereo vision (Intel RealSense D435i) for real-time perception, eliminating reliance on GNSS or pre-defined maps by capturing synchronized RGB and depth data to reconstruct the environment in three dimensions. A perception module processes this data to generate a two-dimensional occupancy grid that identifies traversable regions and obstacles, enabling the A* algorithm to compute optimal paths toward navigation goals. To track these trajectories, a Pure Pursuit controller is employed with an adaptively tuned lookahead distance to ensure smooth path following even on curved segments. A Python-based simulation environment was developed to evaluate performance under realistic agricultural conditions using a simplified stereo SLAM model, achieving effective tracking and obstacle avoidance in row-structured layouts. A physical prototype based on Raspberry Pi 5 is under construction and integration is ongoing as future work, laying the groundwork for robust, vision-based autonomous operation in GPS-denied agricultural scenarios.

Instituciones: Universidad Peruana de Ciencias Aplicadas; Universidad San Ignacio de Loyola