Robots on construction sites are stuck at demonstrator scale, and the bottleneck is validation on real sites rather than the underlying concepts.
That is the core finding of a structured scoping review the European Commission’s Joint Research Centre published, synthesising 129 studies of robotic and automated systems for direct deployment on construction sites and near-site settings.
The review, published in Automation in Construction, was written by Michela Rossi, Cristián Calvo Barentin and Dionysios Bournas. The JRC repository lists the paper as released on November 18, 2026, while the journal dates it to January 2027.
Why care? Because the drivers behind this research are not academic. The study frames the field’s acceleration around labour shortages, safety demands, and productivity and sustainability goals — the same pressures every construction firm is feeling on the ground. A review that maps where the technology actually stands, rather than where roadmaps promise it will be, is a rare thing in this space.
129 studies, one taxonomy
The authors followed the PRISMA-ScR framework — a standardised checklist for scoping reviews that forces transparent selection and reporting of evidence. Out of the corpus, 129 studies made the cut. Each was classified along four axes:
- Operational context of deployment — where on or near the site the system works.
- Functionality — what the machine actually does.
- Team composition — whether robots work alone, alongside humans, or in fleets.
- Lifecycle stage — which phase of a building’s life the system serves.
That multi-dimensional structure is what separates this from a literature dump. It lets a reader see not just what exists, but where the field’s weight sits — and where it conspicuously doesn’t.
Where the robots cluster
The concentration is unmistakable. Three task clusters dominate: additive manufacturing — essentially large-scale 3D printing of building elements — discrete assembly and installation, and perception-driven inspection, where machines use sensors and computer vision to check work that humans would otherwise climb scaffolding to verify.
All three sit in the construction phase. That is the telling part. Operation and end-of-life applications — running and maintaining buildings, then dismantling them — remain sparse across the corpus. The robots are being built for the most visible phase of a structure’s life, and largely ignoring the decades that follow.
Why multi-robot fleets stall in the lab
Multi-robot deployment recurs throughout the 129 studies. Fleets of coordinated machines show up again and again as a concept.
They stay at demonstrator scale. The authors are blunt about the cause, writing that deployment remains “constrained by the difficulty of on-site validation rather than by conceptual appeal.” Translated: nobody doubts the ideas work in principle. Proving them on a live site — with weather, changing geometry, workers moving through the space — is the hard part.
This is the inverse of the factory-automation story. Manufacturing robots thrive because factories are structured, repeatable environments. A construction site is the opposite: unstructured, unique every time, and hostile to the assumptions most autonomous systems are trained on.
Three blockers, same across the board
Across every deployment context the review examined, the same three constraints surface:
- Site variability — the environment shifts constantly, defeating fixed programming.
- Integration effort — bolting a robotic system into an existing construction workflow costs more than the hardware.
- Safety assurance — certifying that a machine can operate safely around human crews remains unsolved at scale.
Safety assurance deserves emphasis. It is not a software bug to patch; it is a certification and liability problem, and it applies whether the robot is printing a wall or inspecting a façade. Until a credible assurance pathway exists, fleet deployment stays in the demo tent.
The research agenda the authors set
The review names three priorities for the field. First, comparable performance and cost reporting — right now, studies measure different things in different ways, which makes it nearly impossible to compare systems or build a business case.
Second, scalable adoption pathways for small and medium-sized enterprises, the firms that make up most of the construction industry but lack the R&D budgets of the few players running pilots. Third, lifecycle expansion toward maintenance and deconstruction, filling the near-empty operation and end-of-life categories.
That second priority may be the most consequential. A technology that only mega-contractors can validate will never reach the density needed to move industry-wide productivity numbers.
Reading it yourself
The paper is a public document and freely accessible through the JRC’s publications repository under handle JRC144021, with the full text also hosted on ScienceDirect. It carries DOI 10.1016/j.autcon.2026.107292 and appears in Automation in Construction, which lists it for the January 2027 issue.
For anyone tracking construction automation — integrators, researchers, or the SMEs wondering when this tech reaches them — the review’s honest accounting of what works, what stays at demonstrator scale, and what nobody is studying yet is more useful than any vendor roadmap.