From Factory to Site

The factory and the robot grew up together, and the factory's whole logic happens to be everything a robot needs. Consider what an industrial robot enjoys on a production line. The environment is structured and controlled: a clean, level, well-lit indoor space, arranged deliberately, that does not change from day to day. The task is repetitive: the robot does the same motion - the same weld, the same pick-and-place - thousands of times, identically. The work is standardised: every part that arrives is the same as the last, positioned the same way, within tight and known tolerances. The robot is fixed in place at a station, bolted down, and crucially, *the work comes to it* on a conveyor rather than it going to the work. And it operates in a bounded, fenced cell, separated from people, so safety can be guaranteed by keeping humans out.

Every one of those conditions plays to a robot's strengths and hides its weaknesses. A robot is superb at repeating a precise motion tirelessly and accurately - so give it the same motion every time. A robot has a limited reach - so bring the work to it. A robot struggles to understand a messy, unpredictable scene - so make the scene clean, known and identical. A robot cannot easily judge whether a human has wandered into its path - so fence it out. The factory is not just a place that happens to contain robots; it is an environment engineered, over decades, to make automation possible, by removing variability and uncertainty until what remains is a task a machine can nail.

This is the single most important lesson to carry forward, and it inverts the usual intuition. We tend to think automation advances by building ever-cleverer robots. In truth, a great deal of automation's success came from structuring the environment and the task so that a relatively simple robot could cope. The factory automated not only because robots got good, but because the work was shaped - standardised parts, fixed stations, repetitive sequences - into something a robot could do. That insight - structure the world, and the robot follows - is the key that unlocks the whole of robotic construction, because it tells you immediately where to look for success: wherever the construction environment can be made more factory-like.

Not everything breaks on the way to a building site; a good deal of the factory's gift travels. The robot itself carries over almost entirely. The actuators, sensors and controllers are the same technology; a six-axis arm that welds car bodies is mechanically much the same as one that could weld steelwork or carry a printing nozzle. The core skills carry over too: precise, tireless, repeatable motion, the ability to follow a path exactly, and the link from a digital model to physical action - the same computer-aided-manufacturing idea that drives a factory tool can drive a construction tool. Where a construction task can be made to resemble a factory task - repetitive, well-defined, with parts in known positions - the robot performs much as it would on a line.

Two moves let builders capture that carry-over deliberately. The first is to bring construction into a factory: prefabrication. If you manufacture walls, floor cassettes, bathroom pods or whole room modules in a controlled plant and then transport and assemble them on site, you get to keep the structured environment where most of the work happens. This is why so much of the real, economically successful robotic construction today is off-site - it is construction reorganised to suit the robot, and it is the single most reliable way the field delivers value now (Module 2.2 is devoted to it). The second move is to make the site itself more structured in pockets: set up a controlled zone, survey it precisely, fix reference points, feed the robot an accurate model, and within that bubble a robot can work almost as it would in a plant. A gantry 3D printer erecting a controlled rig over a slab is, in effect, building a small temporary factory around the work.

So the honest picture is not "robots work in factories but not on sites". It is that robots work wherever the environment is structured enough, and the great strategic question of robotic construction is how much structure you can bring to - or manufacture around - the work. Carry the robot's real strengths with you, engineer as much factory-like order as the situation allows, and a surprising amount becomes possible. The failures come not from the robot but from asking it to work where no such order can be had - which is the subject of what breaks.

Now the hard half. The moment a robot leaves the factory for an ordinary construction site, the conditions that made it succeed are removed one by one, and each removal is a genuine engineering problem rather than a detail. The fixed station breaks: on a site there is no conveyor bringing identical parts to a bolted-down arm. The workpiece - a wall, a floor, a whole building - is enormous and fixed in place, so the robot must travel to the work, over rough ground, and know exactly where it is at every moment. Mobility and localisation, free in a factory, now must be solved.

The known, unchanging world breaks: a site is never the same twice, not even hour to hour. The ground is uneven and shifting, materials are stacked in different places, other trades rearrange everything, and the weather intrudes with heat, rain, wind and dust that a factory never sees. The robot can no longer assume the world matches its program; it must sense and understand a messy, changing scene, which is far harder than executing a known motion. The standardised part breaks: construction is full of one-offs and loose tolerances - a "300 mm" block is roughly 300 mm, a wall is plumb to within a few millimetres over metres, and almost nothing arrives in a precisely known position. A robot tuned to sub-millimetre factory tolerances must now cope with slop and variation that a human absorbs without thinking.

The fenced cell breaks: a site is full of people - other trades working alongside, in the open, unpredictably. You cannot simply fence everyone out, so the robot must be safe to work near humans, which is a profound and still-maturing challenge that Module 7.3 treats as binding. And the repetitive sequence breaks: a building is assembled in a long, interdependent sequence of different tasks that never repeats identically between projects, so there is far less of the sameness that lets a robot pay back its cost through sheer repetition. The pattern is unmistakable and worth stating plainly: it is almost never the robot that breaks - it is the environment. The machine is capable; the site is hostile to the conditions it needs. That realisation is the doorway to the next lesson, where we give this hostile environment its proper name and examine it in full.

Put the carry-over and the breakage together and a clear strategy for robotic construction falls out - one that explains almost everything you will see in the rest of this course. Because robots succeed in proportion to how structured the environment is, the field has three honest routes, and real projects use one or a blend of them.

The first route is move the work to the structure - prefabrication and off-site manufacture. Rather than fight the site, take as much of the building as possible into a factory where the robot's advantages survive intact, then transport and assemble. This is the most mature and commercially successful path today, and it is why "robotic construction" in practice so often means a factory making building components rather than a humanoid on a muddy site. The second route is bring structure to the site - survey it precisely, establish reference points, create a controlled zone, feed the machine an accurate digital model, and operate within that engineered bubble. A 3D concrete printer working from a fixed gantry on a prepared slab is doing exactly this: manufacturing a patch of factory-like order around the work.

The third route is specialise and support - accept that a truly general site-building robot is not near, and instead deploy narrow machines that do one well-bounded task (a single repetitive operation, a survey, a material move) with plenty of human help around them. This is why real on-site construction robots are overwhelmingly specialised and human-supported rather than general and autonomous - a theme Module 2 explores machine by machine. What you will almost never see working reliably is the fantasy version: a general-purpose autonomous robot turned loose on an ordinary, unstructured site to build freely. The honest frontier is structured work, off-site manufacture, and specialised supported machines - and the binding judgements about whether any particular deployment is safe, reliable and economic belong to the engineers, the manufacturers' evidence and the project team, not to the marketing. Keep the factory-to-site lens and you can predict, better than most, which robotic construction claims are plausible and which are theatre.