TALOS proposes, people decide: six technical briefs on smarter solar O&M

Every one of the six briefs contains a walkthrough of an ordinary working day, and that is the best place to start reading. A plant manager opens the platform in the morning. Overnight the system has lined up three jobs: a cleaning run on the dirtiest section, a mowing pass, and an inspection of an area that has been underperforming. The manager approves the cleaning and the inspection, and holds the mowing back until the ground is drier.

Nothing leaves the yard until somebody has said yes. That approval is step four of six, and the method is built around it.

The six steps of the TALOS solution

TALOS describes what it built as a loop, and the loop is the same at all three pilots. Monitor, detect, recommend, validate, execute, report.

The platform watches production, weather and conditions in the field. It locates what is dragging output down and works out what a fix would be worth. Then it stops and waits for a person.

Validation is that human step. The plant manager can move the date, change the location, send a different machine, or set the job aside altogether. Only then does anything go out, and even while a job is running the team can pause it or stop it outright from the same screen.

What is doing the thinking

Underneath the platform sits a central store of plant data: SCADA and meteorological readings, mission logs from every robot that has been out, and a registry of faults. A digital twin runs alongside it, modelling how the plant ought to be performing, so the difference between that and the real readings shows up as a number instead of a hunch.

Detection software flags anomalies at panel, string and inverter level, and attaches a figure to each one in energy and in money. That second figure matters more than it sounds. It lets the recommendations engine rank jobs by what they are actually worth, so a cheap fix that recovers a lot of production rises above an expensive one that recovers very little. What reaches the team is a shortlist with reasons attached, in place of a dashboard to interpret and a calendar to follow.

The screen the team actually uses

Live missions appear on a map, with each machine’s location, battery level and progress, and camera streams where the signal allows. On the Alqueva reservoir that map does a good deal of work, because the 5 MW array sits out on open water and cannot be seen from the office at all.

After every job there is a short feedback form, and what the team writes in it feeds into future recommendations.

Training came with the system: an online introduction, a session on site during the first field demonstration, and a follow-up afterwards. At Randwijk it was designed for growers and agronomists who had never operated a robot in their lives.

The machines, and what each site needed

At Cruz del Hierro, near Ávila, a 56 MW hybrid site with 28 MW of PV, six robotic solutions work under the platform. Three drones carry ordinary and thermal cameras, which pick up hot spots and soiling the eye would miss; they fly pre-planned routes on their own, up to two airborne at once, and land to centimetre accuracy. One of them lives in an automated garage that launches and recovers it with nobody present. A single ground vehicle handles two jobs by reconfiguring itself between a cleaning rig and a mowing rig, swapping over and recharging without help. The third solution is built for scale: a cleaning robot works the rows while a support unit follows behind carrying up to a thousand litres of water, and panels mounted on the system top up its charge as it goes.

Alqueva needed something else entirely.

The floating cleaner does not drive. It runs on rails built into the platform structure, moves from one row to the next automatically, draws water from an on-platform sprinkler system, and charges wirelessly at a docking point on the rails. Getting the inspection drone out to the array is the job of an uncrewed support boat, which means the drone’s charge goes into inspecting panels instead of crossing open water and coming back.

The orchard at Randwijk is the smallest site in the project at 50 kW, and the most crowded. A drone inspects the panels. A ground robot drives between the pear rows carrying a camera-and-sensor viewer that checks the condition of the trees, spots people and obstacles in real time, re-plans its route around them, and counts fruit to support yield forecasts. When both are working the same field, the platform keeps them apart: the ground robot holds back or re-routes while the drone is overhead, then carries on.

The numbers, and what kind of numbers they are

These are project targets rather than audited results, and the briefs are careful to say so. Across all three pilots TALOS is aiming at up to 90% less human exposure to risk, maintenance periods shortened by more than 70%, and around 5% lower O&M costs, with up to 10% less production lost to faults and soiling. At the land-based and floating sites, where panels are washed, the targets add roughly 10% better use of resources in maintenance work and around 35% less water.

What the pilot teams would tell you

The recommendations sections are the least glamorous part of the briefs and probably the most useful.

Supervise autonomy during the early months, because operator validation is what turns a new system into a trusted one. Plan the logistics that autonomy quietly depends on: upkeep, water, somewhere to charge. Update roles, procedures and safety rules, then train people against them.

And treat the platform as the durable part. It is robot-agnostic by design, so a site can swap machines or add new ones and still run the same six steps.

Where to find them

The six briefs are available to download now at talosproject.eu. Briefs 1, 3 and 5 cover the platform at each pilot, written for operators, asset managers and plant owners. Briefs 2, 4 and 6 cover the robots and the software behind them, for robotics providers and technology partners. Each runs to a few pages, with a glossary, a workflow diagram and the partners responsible for each piece.