Spatial & Extrusion Printing

A desktop FDM printer and a giant concrete printer share a limitation few people question: they build planar layers. The nozzle traces a flat slice, the bed (or gantry) drops one layer height, and the next slice goes on top, fused to the layer below. Every strand is supported by material already laid, which is why steep overhangs need sacrificial support and why the whole machine only needs three axes - X, Y and the slow climb in Z.

That planar logic is a choice, not a law of additive manufacturing. Mount the same extruder on a six-axis arm and the nozzle can point and move in any direction. Two freedoms follow. First, non-planar printing: instead of flat slices, the arm can follow curved layers that hug a doubly-curved surface, so the strands run along the form rather than staircasing across it - stronger, smoother, no stepped overhang. Second, the radical one, spatial extrusion: the nozzle draws individual strands directly through space, not as a filled surface at all but as a lattice or scaffold of lines in mid-air, each strand a member of a structure. The arm has turned the printer from a slice-stacker into something closer to a machine that draws in three dimensions.

It is worth being clear that these two freedoms are different in degree. Non-planar printing still makes a continuous, supported skin - it just lets the layers curve, which alone solves the ugly staircasing on domes and shells and improves strength by running material along the surface. Spatial extrusion is the more radical leap: it gives up the continuous surface entirely for a lattice of discrete strands in open space. Most real work sits somewhere on that spectrum, and choosing where is a design decision.

Spatial printing works because a molten strand does not stay molten. As thermoplastic (or metal) leaves a hot nozzle, it is soft; if it cools and stiffens fast enough, it becomes a rigid little beam before it needs anything underneath to hold it up. So the arm can extrude a strand from point A, carry it out into open space, and land it on point B - a bridge that solidified as it was drawn. Chain thousands of these and you get an open lattice: no infill, no supports, mostly air, all structure in the lines.

The governing physics is cooling. Extrude too fast, or in still warm air, and the strand sags or droops before it sets - the print slumps. Cool too aggressively and layers or strands will not fuse, so the part is weak at the joints. Real spatial printing tunes temperature, extrusion rate, travel speed and active cooling together so each strand firms up just in time to support the next. Materials matter: fibre-reinforced polymers stiffen and hold shape better than plain plastic, and metal spatial printing (essentially robotic welding, laying molten metal bead on bead) produces self-supporting steel members - the approach behind robotically 3D-printed steel structures. The design payoff is efficiency: material goes only where structure needs it, following the force flow, so spatial-printed parts can be dramatically lighter than a solid printed block.

Think of it as the difference between a filled surface and a drawn line. Flat-layer printing fills area - it lays down solid skins and infill to make a volume. Spatial extrusion abandons the idea of surface almost entirely and makes structure out of members: each strand is a tiny strut, and the object is a truss or lattice of struts with air in between. That reframes the design task. You are no longer slicing a solid; you are designing a network of lines - where they run, how they meet at nodes, how the load travels from one to the next. It is closer to designing a space-frame than to 3D printing as most people picture it, which is exactly why it appeals to architects: the logic of members-and-nodes is the logic of structure itself.

One of the most elegant uses of spatial extrusion is not to make the final object but to make a mould. In the mesh-mould technique, pioneered at ETH Zurich, a robot spatially prints a dense three-dimensional polymer mesh - a self-supporting cage of strands - and then wet concrete is packed into it. The mesh is fine enough that the concrete does not flow out but is held, so the printed lattice acts as both formwork and reinforcement in one. It collapses two of construction's most labour-intensive trades - building formwork and tying rebar - into a single robotic operation, and because the mesh is printed from a model it can vary freely to make non-standard, load-following concrete walls.

Mesh-mould shows the real pattern of robotic additive in building: not printing the whole thing, but printing the hardest, most bespoke, most labour-heavy part and letting conventional materials do the rest. The same thinking gives us printed connectors that join standard timber or steel, printed nodes for complex space-frames, and printed textures or facades applied to ordinary structure. Structural performance here is genuinely novel and unproven by long track record, so - as always - it is validated by testing and signed off by engineers, never assumed. The design opportunity is to ask which one component is so complex or variable that only a robot drawing in space can make it economically.

In flat-layer printing, slicing is almost automatic: software cuts the model into horizontal layers and fills each one. Spatial printing throws that away. Now the machine needs a path through three-dimensional space - an ordered sequence of strands - and the order matters enormously, because every strand must be reachable and buildable at the moment it is drawn. You cannot lay a strand where the nozzle would crash through structure already printed, and you cannot draw a bridge before its two landing points exist.

So spatial path planning juggles several hard constraints at once: support (is there something for this strand to land on, or has it cooled fast enough to self-support?), collision (can the whole arm, not just the nozzle, reach this pose without hitting the growing part?), reachability (is this pose inside the arm's workspace and within its joint limits?), and sequence (does building order let each strand set before the next relies on it?). This is why spatial printing lives in computational tools - Grasshopper with robot-control add-ons, or frameworks like COMPAS - where the strand network, the robot kinematics and the build sequence are solved together, then simulated before a single gram is extruded. The creativity of drawing structure in mid-air is real, but it is bought with serious planning; the arm gives freedom, and software has to make that freedom buildable.

A useful way to feel the difference: flat-layer slicing is a solved, automatic problem - press a button and the slicer fills every layer - because the constraints are trivial (each layer sits flat on the last). Spatial printing has no such guarantee, so the planning shifts from the machine back to the designer and the algorithm. You decide the strand order the way a bridge-builder decides erection sequence: nothing can be built before what holds it up, nothing can be reached through something already there. Get it wrong and the simulation shows the nozzle plunging through a printed strand, or a bridge drawn to a landing point that does not yet exist. That is why serious spatial printing is always simulated first, printed second - the virtual build catches the impossible orderings so the physical one does not waste material discovering them.