The first results from the TALOS Open Call are in, and they are worth reflecting on. 372 applications from 32 different countries, a clear signal that the topic — using robotics and AI to improve solar plant maintenance — is striking a chord across the European startup ecosystem.
Of the 166 formally submitted applications, 134 were eligible, 123 in scope, 89 above the technical threshold. Twenty teams were invited to the Jury Day. Proposals were ultimately selected. Each received up to €200,000 and nine months working side by side with our technical team, at real operational sites: Villacastín in Spain, Alqueva in Portugal, Randwijk in the Netherlands. No simulations, no labs. Open fields, real panels, real conditions.
Here is an overview of each, with their scope and objectives.
One of the most concrete challenges in managing a large solar plant is aerial inspection. Plants are growing in size and moving into hard-to-reach areas. BVLOS-RAISI (Aerial Tools ApS, Denmark) responded by developing VTOL drones capable of flying autonomously beyond the operator’s visual line of sight. The data is analysed by AI algorithms and returned as automatic anomaly reports.
ARAVI (iKnowHow SA, Greece) tackled a less visible but equally real problem: defects in the backsheet of PV modules — loose connectors, damaged cables, cracks — are hard to detect through conventional inspections, yet can lead to energy losses and safety risks.
Electroluminescence inspections are among the most effective methods for detecting hidden defects in PV modules, but they come with a major constraint: they must be carried out at night, require specialised equipment, and are difficult to scale across large plants. EL-INSPECT (Vertliner P.C., Greece) automated the entire process using a drone. The software automatically maps the GPS coordinates of modules from satellite imagery and sends them to the drone. AI algorithms then analyse the data in the cloud and generate automatic defect reports. The demonstration took place at a plant in Greece which provided similar conditions to TALOS’s site. The goal is to turn EL inspections from an exceptional operation into a routine practice.
Two of the participating proposals explored the same question from different angles: can extended reality reduce the need to send technicians physically on site?
TERRA (Presence XR Solutions S.L., Spain) integrated a VR system onto a ground robot (UGV), allowing one or more users to operate remotely on the plant with SCADA data overlay, maps from satellite imagery and CAD files, and control via voice commands and buttons compatible with ROS2.
CLARO-XR (Mercury Retrograde LDA, Portugal) built something more collaborative: an environment where the on-site technician uses AR while the remote supervisor uses VR, with real-time communication and inverter, string and module data overlaid directly onto the physical components. Position calibration is handled via AprilTags.
ATHINA (Plegma Labs, Greece) installed cameras with automated species recognition around the Villacastín plant to monitor flora and fauna, responding to growing regulatory requirements. During testing, something unexpected emerged: the same system proved useful for early fire detection. Same cameras, different algorithms, a secondary use case that came directly from the field.
Three proposals worked at the agrivoltaic site in Randwijk, in the Netherlands, where solar panels and a pear orchard share the same land. The challenges here are different: agricultural management intertwines with energy production, and decisions on irrigation and harvesting require site-specific microclimate data.
PVGROW (UAB Rediga, Lithuania) installed sensors to monitor microclimate variables under three different panel configurations and built a web platform with an alert and recommendation system for crop management. The system also includes an AI module to estimate the optimal harvest time and detect pear rust.
SUNNY (Amigo S.r.l., Italy) developed an ML model that correlates local sensor data with traditional weather forecasts to generate seasonal forecasts specific to each panel configuration, particularly useful for sites with microclimatic conditions that are hard to predict using standard models.
FRUITURE built a robotic arm with an electro-adhesive gripper to harvest pears without damaging them, replicating the motion used by human pickers.
MooringScout (Indymo B.V., Netherlands) and PERCEIVE (Space Mobility P.C., Greece), which had received an extension until the end of March to complete their work, have now also demonstrated their solutions and received positive feedback from the TALOS team.
The outcome is positive, but above all, it is honest. In light of the challenges addressed and the practical limits encountered, the results point to a genuine market appetite for the kind of technology that TALOS integrates and deploys, and to a broader shift towards automated and sustainable management of photovoltaic systems.
Cover photo by Andreas Gücklhorn on Unsplash
