DEPTHNAV: An autonomous navigation system using depth camera and deep learning in a commercial chicken farming house

Abstract

The increasing automation in commercial livestock farming has raised the demand for more advanced indoor robotic navigation systems. However, current technologies face limitations such as sensitivity to ambient light, inadequate navigation accuracy, and challenges in long-term reliability. In this study, a novel navigation system was proposed, DEPTHNAV (Depth Enhanced Poultry House Navigation), designed specifically for commercial stacked cage farming environments. DEPTHNAV utilizes a depth camera as its primary imaging sensor. By processing pseudo-color images derived from depth data using YOLOv8 (You Only Look Once), the system achieves accurate longitudinal positioning. On the upper floor, with a speed of 0.2m s⁻¹, the standard deviation in longitudinal positioning was less than 1.36$\times {10}^{-2}$ m, while on the ground floor, at a speed below 0.3m s⁻¹, it was under 1.83$\times {10}^{-2}$ m. Additionally, the chassis’ yaw angle was extracted by analyzing point cloud data, ensuring precise lateral control for straight-line movement within aisles. On the upper floor, at 0.2m s⁻¹, the standard deviation of lateral deviation was under 2.21$\times {10}^{-2}$ m, and on the ground floor, it was below 1.80$\times {10}^{-2}$ m at a speed under 0.3m s⁻¹. Its marker-free design simplifies implementation while offering high accuracy and adaptability, making it ideal for scalable applications in automated poultry farming.