Low-cost autonomous unmanned ground vehicle for in-situ phosphorus screening to support zone-based fertilization in maize crops

Villacorta L.; Medrano A.; Lara M. · Computers and Electronics in Agriculture (2026) · DOI: 10.1016/j.compag.2026.112349

Resumen

Conventional fertilization in Peruvian maize relies on generalized, empirical criteria because laboratory soil analysis incurs high cost and provides low spatial resolution. This study presents the design, implementation, and field validation of a low-cost autonomous unmanned ground vehicle (UGV) for in-situ soil sensing for the cultivation of hard yellow maize (Zea mays L.). The platform combines an auger-assisted mechanism that inserts low-cost electrochemical probes up to depths of 15–20 cm into the soil and completes a measurement cycle within 38 s. The sensor-fusion navigation system combines the Global Navigation Satellite System (GNSS) data with visual inertial odometry data. Among the soil parameters acquired, phosphorus was the only channel yielding usable agreement with accredited laboratory references (INIA LABSAF, ISO/IEC 17025); nitrogen and potassium channels showed high volatility and systematic underestimation, respectively. Hence, the N and K channels are reported herein only for transparency. Through Random Forest calibration of the P channel, the raw-probe root mean square error (RMSE) was reduced from 54.2 mg/kg to 25.6 mg/kg under repeated cross-validation (a 53% reduction), with the raw signal significantly correlated with the laboratory reference (r=0.414, p=0.017). However, the cross-validated coefficient of determination remained near-zero (R2≈0.01) at the available sample size. Hence, the calibrated probe serves as a qualitative tool, rather than a quantitative one, for relative P screening. The sensor-fusion strategy achieved a mean positioning error of 1.02 m, a 91% reduction compared to GPS-only navigation, corresponding to open-field waypoint navigation rather than canopy-stage inter-row guidance. A metric named the terrain irregularity index (TII) was determined for the interpretable detection of furrows and obstacles, and georeferenced measurements were transmitted in real time over 4G LTE to a platform that visualizes relative P variability to support zone-based fertilization planning. Overall, this study demonstrates that autonomous low-cost soil monitoring is feasible as a complementary, P-oriented screening tool for zone-based fertilization in smallholder maize. In addition, the present limitations for N and K sensing, quantitative calibration at small sample sizes, and full inter-row autonomy are made explicit.

Instituciones: Universidad Peruana de Ciencias Aplicadas