Robotic Arm with Artificial Intelligence for Subsurface Irrigation of Onion in Urban Gardens of Lima

Huachaca R.A.; Serquen W.A.P.; Acosta A.K.N.; Navarro J.A.C.; Vilcarromero J.J.F.; Malpartida E.A.S. · 2026 12th International Conference on Mechatronics and Robotics Engineering, ICMRE 2026 (2026) · DOI: 10.1109/icmre69538.2026.11533917

Resumen

Urban community gardens in arid cities such as Lima often irrigate uniformly over the soil surface, ignoring the actual root zone and wasting water. This paper presents a vision guided robotic subsurface irrigation system that combines an RGB camera, a Raspberry Pi and a four degree of freedom manipulator equipped with a water injector. A U Net Small network segments the wetted bulb from red onion images; trained on one hundred and seventy six samples, it achieved an intersection over union score of about zero point sixty five and an F1 score of about zero point seventy four, sufficient to localize the active root region. On top of this perception layer, three visual servoing schemes are implemented and compared: image based, pose based and a hybrid controller. The pose based scheme provided the best tracking performance, with complete convergence in roughly four seconds and negligible residual position error. Using standard crop coefficients for onion, the automated subsurface strategy yielded a theoretical application efficiency between ninety two and ninety five percent and an estimated reduction of up to fifty percent in water use compared with gravity irrigation, applying between fifteen and twenty millilitres per plant directly in the root zone. These results indicate that the proposed architecture is a promising low cost platform for precision irrigation in urban gardens of shallow rooted crops such as onion.

Instituciones: Universidad Peruana de Ciencias Aplicadas