PREVID Operational Group: Advanced digitization and agronomic prediction in table grapes using quadruped robotics, artificial vision and artificial intelligence

PreVid addresses a strategic need in the sector: to digitize, automate, and objectify critical agronomic and commercial management processes in a high-value production system like vineyards. Dividing the project into four specific objectives allows for a structured and progressive development of the system. A series of expected results for end users have been identified: substantial improvement in the objectivity and accuracy of bunch counting and production estimation; reduction of costs associated with manual sampling and field technicians; optimization of agronomic management through decisions based on predictive data; improvement in logistical and commercial planning, especially for export campaigns; access to a scalable tool, with the possibility of replication in other vineyard structures or similar crops; and generation of knowledge on intra-plot variability, factors limiting vigor, and phenological prediction, with strategic value for the digitalization of the crop.

Result 1. Initial prototype of the quadruped robot adapted to trellises. Activity 1. Definition of agronomic, technical, and operational environment requirements (AGERPIX, TECNOVA, MOYCA, CODESIAN); Develop and validate an autonomous terrestrial robotic solution equipped with Agerpix OnFRUIT cameras for counting grape bunches, assessing vigor, estimating quality, and predicting the productive state in trellised vineyards, thus improving efficiency in agronomic management, harvest forecasting, and classification for export. In this initial phase, the specific agronomic, technical, and operational needs for open-air table grape viticulture will be gathered. The robot's requirements in terms of movement and autonomy, the necessary machine vision systems and sensors, as well as the work scenarios and environmental conditions that the robotic solution must withstand, will be defined. This will establish a common working basis for the subsequent design, development, and validation of the system.

Result 1. Initial prototype of the quadruped robot adapted to a trellis. Activity 2. Design and adaptation of the quadruped robot to vineyard conditions under a trellis (AGERPIX, TECNOVA, MOYCA). In this phase, the technical design and adaptation of the terrestrial quadruped robot will be carried out for its movement and autonomous operation in open-air vineyards under a trellis. The specifications defined in Activity 1 will be addressed, adapting the structure and the mobility, energy autonomy, and communication systems to operate on agricultural terrain with irregularities, slopes, roots, and low environmental exposure. In addition, the supports and placements of the overhead cameras and onboard sensors will be defined and validated to ensure their correct operation under varying lighting conditions.

Result 1. Initial prototype of the quadruped robot adapted to a vineyard. Activity 3. Development of communication and autonomy for the field navigation robot. (TECNOVA) This activity will address the development and implementation of the quadruped robot's autonomous navigation and real-time communication systems for its subsequent deployment in the field. Tecnova will be responsible for adapting and configuring the autonomous control, secure communication, and remote monitoring systems, enabling the robot to reliably move between the vineyard rows, avoid obstacles, and maintain a stable connection with the data management and operations control platform. This activity is key to ensuring the system's continuous operation in real agricultural environments with varying coverage and changing environmental conditions.

Result 2. Functional prototype with integrated sensors and machine vision. Activity 1. Integration of agronomic cameras and sensors into the robot (AGERPIX, TECNOVA, MOYCA). The objective of this activity is to functionally and operationally integrate machine vision systems and agronomic sensors into the adapted quadruped robot, ensuring their correct operation and synchronization with the navigation and communication systems. The optimal locations for overhead cameras and sensors will be defined, the mounts will be adapted, and the data and power connections will be established. The first joint operation tests will be carried out in a simulated environment. Data will be obtained from farm sensors to validate correct operation under real-world field conditions.

Result 2. Functional prototype with integrated sensors and machine vision. Activity 2: Development of machine vision algorithms for counting, bunch size, and vegetative vigor. (AGERPIX, CODESIAN). This activity aims to develop and implement machine vision algorithms capable of detecting, counting, and estimating the size of grape bunches in a vineyard, as well as evaluating the vegetative vigor of the plants from images captured by the overhead cameras integrated into the robot. The algorithms must be robust under varying lighting conditions, bunch inclination, and agronomic heterogeneity.

Result 3. Operational AiCrop platform with integrated data from the robotic system. Activity 1. Development of autonomous adaptive navigation routines for AiCrop based on terrain conditions and crop structure. (TECNOVA, CODESIAN). This activity aims to develop and implement autonomous adaptive navigation routines for the quadruped robot, capable of modifying its movement behavior according to terrain conditions (hard ground, irregularities, roots, slopes) and the trellis crop structures. These routines will be integrated into AiCrop for remote parameterization, navigation incident reporting, and route optimization on the farm.

Result 3. Operational AiCrop platform with integrated data from the robotic system. Activity 2. Data integration in AiCrop and development of predictive models (CODESIAN, MOYCA). The aim is to integrate all the data collected by the robot (bunch count, size estimation, vegetative vigor, climatic and soil conditions, navigation incidents, etc.) into the AiCrop platform and develop predictive models for harvest and exportable quality, as well as vigor maps of vineyard plots. This activity will transform raw data into valuable information for agronomic and commercial decision-making.

Result 3. Operational AiCrop platform with integrated data from the robotic system. Activity 3. Generation of decision rules and agronomic recommendations based on AI. (CODESIAN, MOYCA, TECNOVA). Develop and implement a system of decision rules and agronomic recommendations, generated from data collected by the robot and processed in AiCrop, that allows farm technicians to optimize the management of the trellised vineyard. These recommendations will be based on predictive models and the automated interpretation of variables such as vigor, bunch count, harvest prediction, and exportable quality.

Result 4. Agronomic validation of the integrated system in a functional operating environment. Activity 1. Definition of agronomic validation scenarios and planning of operational campaigns (MOYCA, TECNOVA). Define the real-world farm scenarios where the robotic solution will be validated, establishing the agronomic conditions, plot types, test schedules, and evaluation procedures. Additionally, plan the field validation campaigns, ensuring adequate coverage of the different growing conditions, phenological stages, and operational situations that may be encountered in trellised vineyards.

Result 4. Agronomic validation of the integrated system in a functional operating environment. Activity 2. Design and deployment of the virtual test environment for verifying data integration in AiCrop (TECNOVA, AGERPIX, CODESIAN). Develop a simulated virtual environment where controlled tests of data integration from the robot's cameras, sensors, and navigation system can be performed with the AiCrop platform. This environment will allow for the prior validation of the connectivity, compatibility, and behavior of predictive models and machine vision algorithms without the need for field tests, optimizing time and error debugging.

Result 4. Agronomic validation of the integrated system in a functional operating environment. Activity 3. System validation in MOYCA pilot plots. (MOYCA, TECNOVA, AGERPIX, CODESIAN). Verify and validate, under real field conditions, the integrated operation of the quadruped robot, the artificial vision systems and agronomic sensors, real-time communication, and the AiCrop platform. This activity will allow for the evaluation of the operability of the counting, vigor, and quality algorithms, the predictive models, and the AI-based agronomic recommendations, and for adjusting for any issues that may arise under production conditions.

RD1: Dissemination Plan. Effective Internal Communication. Act1.RD1: Design of the Dissemination and Communication Plan: A strategic plan will be developed that defines the project's visual identity, key messages, target audiences, communication channels, and responsibilities among GO members. This document will include guidelines to ensure consistent communication throughout all phases of the project and alignment with the visibility objectives established in the call for proposals. In addition, basic materials such as a logo, templates, posters, roll-up banners, and informational brochures will be designed.

RD1: Dissemination Plan. Effective Internal Communication. Act2.RD1: Internal Coordination and Communication Monitoring: Mechanisms will be implemented to foster effective internal communication among partners through digital collaboration tools (Teams, Google Drive, etc.), regular meetings, and shared protocols. A system for monitoring and tracking communication activities will be established to ensure traceability and detect any deviations for corrective action.

RD2: Dissemination and technology transfer through participation in events and project-related workshops. Act1.RD2: Participation in industry events and trade fairs: GO members will attend agri-food trade fairs, specialized conferences, and innovation forums to present project progress through technical presentations, booths, and informational materials. This participation will expand the network of contacts, validate approaches with the sector, and disseminate the potential of the developed system. Planned events include Fruit Attraction, FAME INNOWA, InfoAgro Exhibition, and Science Week, among others.

RD2: Dissemination and technology transfer through participation in events and the organization of project-related workshops. Act2.RD2: Organization of project-related dissemination workshops: At least two workshops will be organized during the project (one mid-term and one final), aimed at producers, technicians, students, innovation platforms, and policymakers. The workshops will include technical presentations, field demonstrations, and roundtables to foster dialogue among key stakeholders in the wine and technology sectors.

RD2: Dissemination and technology transfer through participation in events and project-specific workshops. Act3.RD2: Cooperation with other Operational Groups and stakeholders: Exchange opportunities with other national and international Operational Groups will be fostered through networking events, joint webinars, and participation in transfer workshops organized by networks such as the RRN or regional platforms. These actions will allow for the sharing of experiences, alignment of methodologies, and the scalability of results.

RD3: Project dissemination through virtual activities. Act1.RD3: Creation of the official project website: A multilingual website (Spanish/English) will be developed that will include the objectives, participants, activities, progress, and results of the GO. This website will act as a document repository and as a visibility and transparency tool, with access to downloadable materials, news, and sections for different audiences.

RD3: Project dissemination through virtual activities. Act2.RD3: Website and social media updates: Profiles will be created on specialized social media platforms (LinkedIn, X/Twitter, YouTube), and at least three electronic newsletters will be published throughout the project: a launch newsletter, an interim progress update, and a closing newsletter. Content will be tailored to different target audiences and will include news, use cases, field testimonials, and upcoming events.

RD3: Project dissemination through virtual activities. Act3 and Act4. RD3: Production of audiovisual content (Newsletters and audiovisual material): At least three videos will be produced reflecting different stages of the project (beginning, field implementation, final results). These resources will include animations, interviews, and recordings of workshops, and will be used both in person and on the web and social media. They will be complemented by interactive infographics, presentations, and downloadable visual resources.