The GreenBot autonomous system completes successful field trials for sustainable woody crop protection

El proyecto GreenBot da un paso clave hacia la agricultura sostenible con la puesta en campo de un vehículo autónomo de alta precisión diseñado para el control inteligente y localizado de malas hierbas en cultivos leñosos como almendro, cítrico y olivar. Este avance, fruto de la colaboración público-privada, integra inteligencia artificial (IA), robótica y visión por computador para optimizar el uso de productos fitosanitarios, reducir costes y mitigar el impacto ambiental de la agricultura intensiva.

Un reto agronómico con respuesta tecnológica

Las malas hierbas representan una amenaza constante para la producción agrícola, con pérdidas estimadas de hasta un 40 % del rendimiento en cultivos. Los métodos convencionales de control, basados en la aplicación generalizada de herbicidas, no solo son costosos (hasta un 30 % del coste de producción) sino también perjudiciales para el medio ambiente debido a fenómenos de deriva o escorrentía.

GreenBot aborda este problema mediante un enfoque preciso y dirigido, adaptado al entorno complejo de los cultivos leñosos, donde el acceso bajo la copa de los árboles y la presencia de sistemas de riego hacen inviable el uso de maquinaria convencional sin riesgo de daño.

Resultados preliminares y validación en campo

The GreenBot project has taken a key step towards sustainable agriculture with the field deployment of a high-precision autonomous vehicle designed for the smart and localized control of weeds in woody crops such as almond, citrus, and olive trees. This breakthrough, made possible thanks to a public-private partnership, integrates artificial intelligence (AI), robotics, and machine vision to optimize the use of plant protection products, reduce costs, and mitigate the environmental impact of intensive farming.

A tech response to an agronomic challenge

Weeds pose a constant threat to agricultural production, with estimated crop yield losses of up to 40%. Conventional control methods, based on the widespread application of herbicides, are not only costly (up to 30% of production costs) but also harmful to the environment due to drift or runoff.

GreenBot addresses this problem through a precise and targeted approach, adapted to the complex environment of woody crops, where access under the tree canopy and the presence of irrigation systems make the use of conventional machinery unfeasible without risk of damage.

Preliminary results and field validation

During field tests, the autonomous system proved effective under different light, soil, and plant cover conditions. Areas of improvement have been identified in relation to the detection of small plants in shaded conditions, which has prompted further training of the model with enriched data.

With an inference frequency of 1 second per image, the system is able to operate in real time, without the need for external servers, and has achieved a complete integration between perception, navigation, and localized application, validated by all the technical teams involved.

GreenBot involves a multidisciplinary consortium made up of the University of Seville’s AGR-278 “Smart Biosystems Laboratory” research group, GMV, TEPRO, PIONEER HiBred Spain SL, and Cooperativas Agroalimentarias de Andalucía. The Greenbot Task Force project was scheduled to last 21 months and concluded on 30 June 2025. 

Cutting-edge technology for localized application

As part of this project, GMV has developed an autonomous robotic platform controlled by its uPathWay solution, combining machine vision, smart navigation, and a localized application system for plant protection products. The robot’s features include:

• Autonomous inter-row navigation based on ROS2, GNSS RTK sensors, IMUs, and, optionally, LiDAR or proximity sensors.
• A semi-circular robotic arm that encircles the trunks without stopping forward movement, equipped with spray nozzles that are only activated on the specific area where weeds are detected, minimizing the use of chemicals.
• The system makes it possible to identify the critical area to be treated - between the trunk and the drip line - with great precision, avoiding damage to the crop and ensuring effective intervention on existing weeds.

The weed detection core, developed by the University of Seville, is based on a ZED 2i stereo vision system installed at low height, connected to a 64 GB Jetson AGX Orin processor. An ad hoc trained YOLO-based detection model processes high-resolution images in real time, identifying the species, position, and dimensions of each weed with a spatial accuracy of ±2 cm.

Each detection is converted into a structured dataset (annotated image, class, confidence, 3D coordinates, etc.) that is automatically integrated into the robot’s control and processing system through a REST API implemented with FastAPI.

This project is funded by the 2022 round of grants for European Innovation Partnership (EIP) Operational Groups, within the framework of Rural Development Program of Andalusia 2014-2022, which in turn is covered by the Spanish Ministry of Agriculture, Livestock, Fisheries, and Sustainable Development’s Order of 7 July 2020 (sub-measure 16.1, operations 16.1.2 and 16.1.3).