Robotic Assembly

Every assembly robot, however sophisticated, is built on one motion: pick-and-place. The arm's end-effector - now a gripper, a vacuum cup, a magnet or a custom clamp - closes on a part at a known pick location, the arm carries it along a planned path, and it releases the part at a computed place position and orientation. In industry this is the most common robot job in the world, sorting and packing millions of items an hour.

What makes it powerful for construction is precision under variation. A human bricklayer working to a standard bond places every brick the same way by muscle memory; ask them to give each course a slightly different angle to make a curving, patterned wall and the job becomes slow and error-prone. A robot does not care. It reads a different target for every single element - position X, Y, Z and rotation - from the digital model, and places each one just as accurately as the last. The repeatability that limits nothing here (unlike milling, there is no cutting force to flex the arm) is exactly the strength: sub-millimetre placement, element after element.

The hard parts are the ones humans do without thinking: gripping an irregular part reliably, knowing where the part actually is (feeders, jigs and increasingly vision systems solve this), and planning a collision-free path to set it down among the parts already placed. Assembly is less about brute force than about spatial bookkeeping - and that is precisely what software is good at.

It helps to see why placement suits a robot so much better than cutting does. In milling, the tool pushes hard against solid material, and that force flexes the arm - the very compliance that hurts a robot's accuracy. In pick-and-place there is no cutting force: the arm carries a part and sets it down, so its weakness never gets exercised and its greatest strength, repeatability, comes fully into play. An arm that returns to the same taught point within a tenth of a millimetre, over and over, tirelessly, is close to ideal for placing parts. The machine that struggles to mill steel accurately can position a brick beautifully. That asymmetry is the single best reason to expect assembly, not cutting or printing, to be where robots first earn their keep in real construction.

The landmark demonstration was masonry. Gramazio Kohler's early work took an ordinary brick - the oldest modular building unit there is - and let a robot place each one at a computed angle and offset. The result was walls no mason would build by hand: gently twisting surfaces, walls that read as woven or perforated, patterns where the rotation of each brick modulates light and shadow across a facade. Because the robot places from a model, a wall of two thousand all-different bricks costs no more effort to lay than two thousand identical ones - the file simply carries two thousand targets.

Mechanically, robotic bricklaying is pick-and-place plus adhesive: the arm picks a brick, often dips or receives a bead of glue or mortar, and sets it precisely onto the course below. Prefabricated in a shop, whole wall panels arrive on site ready to hang. Commercial systems now also lay brick on site at pace for standard walls, taking the repetitive lifting off human masons. Two lessons matter here. First, robots make differentiation free, so the design value is in geometry a hand cannot economically achieve. Second, real masonry needs the mortar to cure and the wall to stand and carry load - structural adequacy is always an engineer's sign-off, robot or not; the arm guarantees position, not that the wall is sound.

Bricks are small and forgiving; timber scales the idea up. The ICD in Stuttgart and others have used arms to stack, position and fasten timber elements - laths, slats, beams, plates - into structures a carpenter could not economically set out by hand. A robot holds each stick at the exact computed angle while a fastener goes in, or nails and glues layered assemblies where every layer is slightly rotated, building up differentiated timber plates and shells. The same logic reaches steel nodes, precast blocks and dry-stacked components.

With bigger parts, three things change. Grippers get bespoke: a vacuum cup that lifts a brick will not hold a two-metre beam, so end-effectors are designed per element, sometimes with sensors to confirm the grip. Payload and reach become real limits - an arm has a rated payload (a few kilograms to a tonne-plus for the biggest) and a finite reach, so heavy or long members may need a gantry-mounted robot, a robot on a track, or two arms sharing the load (the next lesson). And tolerance stacks up: place forty parts each a fraction of a millimetre off and the errors accumulate, so real cells use jigs, datums and sometimes vision to re-reference against the growing structure. Assembly, unlike cutting, is where the robot's precision and tirelessness translate most directly into buildings.

A further shift with larger parts is the growing need for sensing. When elements are small and delivered by a precise feeder, the robot can place blind - it trusts that the part is exactly where the program says. As parts get bigger, heavier and less perfectly presented, that trust breaks down: a beam may sit a few millimetres off in the rack, or the structure may settle slightly under its own weight as it grows. So advanced cells add vision (cameras that locate the part and the work), force sensing (feeling when a part seats or jams), and touch probing to re-reference against reality rather than the ideal model. This is the difference between a robot that follows a script and one that closes the loop - measuring, comparing, and correcting. Construction, being full of real-world variation, pushes assembly robotics steadily toward that sensor-rich, adaptive end.

Most assembly robots today live in fenced cells: a fast, powerful industrial arm carrying a heavy part is dangerous, so it works behind guarding with interlocks that stop it the instant a gate opens. But the frontier of construction is human-robot collaboration - people and machines sharing a task, because buildings are big, messy and unpredictable in ways a fully-automated line is not.

Two models are emerging. In one, a collaborative robot (a cobot) is built to work near people: force-limited, speed-limited, with sensors that stop it on contact, so a person can hand it parts or work beside it. Cobots trade speed and payload for that safety. In the other, the human and a full industrial robot share a job but not the same instant of space - the robot places heavy elements while a person does the dexterous fastening, choreographed so they are never in the same zone at once. The division of labour plays to each side: the robot brings tireless precision and lifting, the human brings judgement, dexterity and dealing with the unexpected. This partnership - not full automation - is the realistic near future of robots on the building site, and it only works when safety is engineered in from the start, never improvised.

Economically, the case for assembly robotics is strongest wherever the work is repetitive, precise and physically punishing at once - lifting and setting heavy units to exact positions all day. Those are the tasks that injure human bodies and drift in quality as people tire, and they are exactly what a tireless, sub-millimetre arm does best. Freed from them, skilled workers move up to the judgement-rich work machines cannot do. Seen this way, the robot is less a replacement than a power tool at building scale.