One of FlexiGroBots main outcomes is the creation of an AI and data architecture that enables the design, implementation, training, testing and serving of AI models for Agriculture. To achieve that, FlexiGroBots has set up a Kubernetes cluster, where several technologies are deployed: Kubeflow to develop and deploy AI pipelines, MLflow to register AI models, MinIO to store the data and Kserve to deploy AI models in production. Additionally, the infrastructure for a Data Space for Agriculture has been implemented based on IDSA components, providing connectors to interact with the Data Space. Finally, a graphical web interface has been implemented to improve user experience.
In FlexiGroBots, 10 AI horizontal applications have been developed, targeting features as people detection, tracking, and action recognition, which are vital for ensuring safety in human-robot interactions; object detection, disease management, anonymization for privacy compliance, automatic dataset generation to improve AI model training, 3D virtualization for enhanced agricultural analysis, and specific tools for fruit disease detection, pest detection, and weed management. These advancements collectively contribute to the safety, efficiency, and sustainability of agricultural practices.
Finally, the MCC has been implemented, providing support for both UGVs and UAVs, since it allows control and management of the autonomous vehicles. A graphical interface has been designed to ease the use of this tool.
Regarding the pilots, apart from completing the implementation of the use cases, in the last half of the project the efforts have been directed at building a superscenario for each, so that FlexiGroBots solutions could be tested in multi-robot environments.
In pilot 1 (grapevines-Spain), UGVs were used to support harvesting and field monitoring, while UAVs were used to collect high-resolution images. Their superscenario consisted of a fleet of robots equipped with sensors and cameras and connected to a wireless network. Their work is orchestrated by a Fleet Manager within the MCC, and the information is presented to the farmer through a web platform.
Pilot 2 (rapeseed-Finland) supported activities like silage harvesting, Rumex weeding, and pest management. UAVs have also been used, for example, following and tracking autonomous tractors in the field. In their superscenario, 9 UAVs and UGVs have worked collaboratively.
In pilot 3 (raspberry-Serbia and Lithuania), a set of soil sampling robots has been deployed to make a chemical analysis of the soil. The system has also a sprayer used to add chemicals to the plants. In their superscenario, the integration of up to 3 robots working collaboratively in a real environment was successfully tested.
It should be noted that all the KPIs have been reached, so FlexiGroBots can be considered a success. All the results have been actively communicated (scientific publications, social networks, press notes, website). Additionally, a set of virtual demonstrations were held to present the solutions to DIHs and the related community to increase FlexiGroBots visibility and impact and pave the way for the exploitation of all the solutions.
