Robotic Milling & Cutting

Strip away the mystique and a milling robot is a familiar thing: an industrial six-axis arm - a KUKA, an ABB, a Staubli - with a cutting spindle bolted on as its end-effector instead of a gripper. It reads the same kind of instructions as any CNC machine (positions, feed rates, spindle speeds), generated by CAM software and posted to the robot controller. The material is removed exactly as on a router: an end mill spins at speed while the tool follows a toolpath, taking off chips pass by pass until the block becomes the part.

What changes is the geometry of access. A three-axis gantry router moves the tool in X, Y and Z but cannot tilt it; the cutter points down and stays down. A five-axis machine adds two rotary axes and can tilt within a limited cone. A six-axis robot can bring the tool in at essentially any orientation its joints allow, and its arm can reach around a part rather than only down onto it. That freedom is why robots dominate the milling of large, sculptural, deeply-curved forms.

The price is rigidity. A gantry is a stiff steel box; a robot is a chain of six motorised joints, and every joint has a little compliance. Push a cutter hard into steel and the arm deflects - fractions of a millimetre, but enough to matter. So robotic milling lives happily in soft and medium materials - rigid foam, timber, plaster, soft stone, some composites and plastics - and treats hard metals with caution, light cuts and slow feeds.

The single most common robotic-milling job in architecture is rigid foam - EPS and the denser XPS. It cuts like butter, so the arm can take huge, fast passes, and the result is a full-size mould. Cast concrete or spray a composite skin against a milled-foam mould and you get doubly-curved panels - GRC facade elements, seating, monuments - at a cost that would be impossible with hand-built formwork. A designer roughs the block with a large ball-nose cutter at a coarse stepover (the sideways bite between passes, maybe 10-15 mm) to remove bulk fast, then finishes with a fine stepover of 1-2 mm for a smooth ruled or free surface.

Timber is next. Robots carve joinery, mould-like patterns, relief panels and sculptural columns; the arm angle lets a cutter follow the grain of a twisting form or reach into a deep coffer. Hardwood needs sensible feeds and a rigid setup, because here the arm's compliance shows. Stone - marble, limestone, sandstone - is milled with diamond tooling and water cooling; it is slow and abrasive, but robots have revived architectural stone carving, roughing a block from a scan or model before a mason finishes by hand. Across all three the mindset is the same: rough pass to remove bulk, finish pass for surface, and let the six axes reach the angles a router cannot.

The tooling logic is worth internalising because it governs both time and quality. A ball-nose cutter leaves scallops - little ridges between passes whose height depends directly on the stepover; halve the stepover and you roughly halve the scallop height, but you double the number of passes and therefore the time. So finishing a large foam mould to a glass-smooth surface can take many hours even though roughing took minutes. Feed rate (how fast the tool travels through material) and spindle speed are matched to the material: foam tolerates aggressive feeds, hardwood and stone demand restraint. And because the arm can angle the tool, you can keep the cutter's most effective cutting zone on the surface even on steep walls, where a straight-down router would be dragging its tip and burning the finish. This is the quiet craft of robotic milling - not the exotic robot, but the same speeds-and-feeds and roughing-and-finishing discipline every machinist knows, now applied through six axes of freedom.

Milling foam removes material as dust and chips - clouds of it. There is a faster, cleaner way to shape EPS: hot-wire cutting. Instead of a spindle, the end-effector carries a taut wire (or a rigid hot blade) heated electrically to a few hundred degrees. It does not cut - it melts a thin slot through the foam as the arm sweeps it. A single pass can slice through a metre of polystyrene in seconds, with almost no debris and a smooth, sealed face.

The geometric catch: a straight wire can only sweep a ruled surface - a surface you could make by moving a straight line through space. Planes, cones, cylinders, hyperbolic paraboloids and gentle twists are all ruled and cut beautifully; a bulging double curve is not, and needs milling instead. This is exactly why hot-wire robots are the darling of concrete formwork: many elegant column, beam and panel moulds are ruled surfaces, and an arm can carve a whole EPS formwork block in a fraction of the time a mill would take, then the block is cast against and stripped away. A hot blade - an elastically bent, heated strip - extends this to smooth, tangent-continuous curved cuts, a technique developed at ETH Zurich for exactly this formwork use. The wire has a kerf too: the melted slot is a millimetre or two wide, and you offset the path to keep the finished face on size.

Here is the sober part. A good gantry CNC holds a few hundredths of a millimetre across its bed. A six-axis milling robot, out of the box, is far less accurate - its absolute accuracy (how close it gets to a commanded point in space) can be a millimetre or more, even though its repeatability (returning to the same point again) is excellent, often under a tenth of a millimetre. The difference comes from the arm's kinematics: small errors in six joint angles compound along a long lever, and cutting forces flex the structure.

Labs fight this in three ways: calibration (measuring the real arm and correcting its model), stiff, shallow toolpaths (small radial and axial cuts so forces stay low), and external metrology - a laser tracker watching the tool and feeding corrections back. For most architectural work the honest answer is simpler: choose robotic milling where a millimetre is fine (foam moulds, sculptural timber, rough stone) and reach for a gantry or a dedicated 5-axis machine where tenths matter (precision joinery, metal, mould tools). And treat safety seriously - a spinning cutter or a heated wire on a fast, powerful arm is dangerous; robotic cells are fenced and interlocked, and you never share the workspace with a running industrial arm without proper guarding and training.

There is one more honest limit worth naming: the workspace is not a tidy box. A gantry's reachable volume is a simple rectangular prism, easy to reason about. An arm's is an awkward, roughly spherical shell with dead zones - places it physically cannot reach, and singularities, special poses where the arm loses a degree of freedom and can lurch or stall. Good CAM for robots simulates the whole arm through every move to catch reach limits, self-collisions and singularities before the spindle ever spins. So the workflow gains a step the gantry never needed: you do not just generate a toolpath, you verify that this particular arm, from this particular mounting, can actually execute it safely end to end. That verification is where a lot of robotic-milling expertise really lives.