Renewables
New image dataset supports AI navigation for solar panel cleaning robots
An international research team has released an open-access image dataset designed to improve visual navigation for autonomous photovoltaic cleaning robots under various soiling conditions.
Key points
- The dataset contains 6,200 annotated images of solar panels, including clean modules, synthetically soiled panels, and laboratory-tested physically soiled units.
- Researchers from institutions in Malaysia, Canada, Belgium, Norway, and India collaborated on the study, published in Data in Brief.
- The project focuses on identifying structural reference lines to help robotic vision algorithms maintain orientation despite surface contamination.
What happened
According to pv magazine, an international group of researchers has developed an open-access image collection to aid the visual navigation of autonomous solar panel cleaning robots. The dataset comprises 6,200 annotated images, capturing clean panels alongside synthetically generated contamination across five severity levels and physically soiled modules tested in a laboratory.
To gather the visuals, the team mounted a camera on the front of a cleaning robot positioned approximately 15 cm above the module surface. The manual annotation process covered 800 clean images to identify three distinct structural reference lines: white cell-gap lines, as well as horizontal and vertical frame-cell border lines.
Why it matters
Corresponding author Kumeresan A. Danapalasingam told pv magazine that the initiative prioritises visual navigation for cleaning robots rather than conventional panel defect inspection. By maintaining consistent reference-line geometry across various simulation levels, the dataset offers a systematic approach to test whether robotic vision algorithms can detect navigation cues on contaminated surfaces.
Co-author Xin Li noted that different types of contamination obscure reference lines while keeping the underlying geometry intact. This separation of visual degradation from viewing geometry helps researchers understand the failure points of current navigation algorithms and build greater robustness.
What comes next
The academics intend to transition from dataset creation toward achieving reliable autonomous navigation. Their ongoing work involves examining lightweight artificial intelligence models to estimate robot headings relative to panel reference lines, alongside localisation methods that merge visual data with inertial and wheel-encoder measurements for real-world deployment.
Original reporting
This article draws on the following reports. Read them in full on the publishers’ sites.
Context from CEII
Renewables are the fastest-growing part of the world’s energy system. The COP28 goal is to triple capacity by 2030, and solar mini-grids are the fastest route to power for millions of people without electricity.
27.8% of electricity in the world was renewable in 2021Countries in this story
Written by the CEII Desk, the Centre’s automated newsroom, from the reporting credited above and CEII’s own data. Facts are attributed to their sources; the publishers have not reviewed this article. Spotted an error? Tell our editors.
