Total Fabrication: Liquid & Robotic Systems

• Total Fabrication is a manufacturing paradigm that forms structures via entire-volume liquid shaping or integrated robotic construction, avoiding voxel-by-voxel processes.
• LiquiFab employs surface tension under weightless conditions to achieve rapid and scalable production by constraining a liquid polymer within minimal boundary sets.
• Robotic systems like In situ Fabricator integrate precise localization, optimal control, and a seamless CAD-to-fabrication workflow for full-scale, high-accuracy on-site construction.

Total Fabrication denotes a manufacturing conception in which form generation is achieved without voxelwise visitation of every point in space and without the fragmentation of design, placement, and assembly into disconnected stages. In the LiquiFab formulation, Total Fabrication is realized by dispensing an entire volume of liquid polymer into a sculpted set of boundary constraints and allowing surface-tension forces under weightlessness to drive that volume to its final, minimum-energy shape “all in one go,” after which the part is cured and reused as a boundary for subsequent elements (Hochman et al., 19 Dec 2025). In the In situ Fabricator formulation, the term is associated with end-to-end, full-scale digital construction on site, in which a self-contained mobile manipulator closes the loop from CAD to on-site material placement through integrated localization, motion planning, and control (Giftthaler et al., 2017). Taken together, these usages situate Total Fabrication at the intersection of global form-finding, sequential assembly, and full-scale digital production.

1. Conceptual position relative to subtractive and additive manufacturing

In the LiquiFab account, existing digital manufacturing methods are divided into subtractive approaches, where material is removed from a bulk to reveal the desired form, and additive methods, in which material is introduced voxel-by-voxel to create an object (Hochman et al., 19 Dec 2025). Subtractive processes such as CNC milling and laser cutting start from a bulk and remove material to reveal a form; every point that should not remain must be visited and removed, and waste is intrinsic. Layered additive methods such as FDM, SLA, and DLP build an object voxel-by-voxel or layer-by-layer; as the object grows, every new layer or voxel must be addressed by a toolpath or light exposure, so build time grows roughly in proportion to object volume. Volumetric additive methods, exemplified by tomographic VAM, cure an entire volume simultaneously via light patterns, but are limited by optical penetration depth, resin choice, and typically waste uncured resin (Hochman et al., 19 Dec 2025).

LiquiFab is described as fundamentally different from both subtractive and additive manufacturing because it neither removes material from a bulk nor builds by layers or voxels. Instead, one single liquid volume instantaneously assumes its final shape according to the geometry of a small set of bounding surfaces; no material is discarded and no voxel-by-voxel scan is required. Once the liquid wets its constraints, surface-energy minimization performs the shaping all at once (Hochman et al., 19 Dec 2025).

The In situ Fabricator literature addresses a different scale and operational context. There, Total Fabrication refers to a self-contained, all-terrain mobile manipulator whose mechanical design and hardware enable true 1:1 on-site construction, and whose software stack closes the loop from CAD to on-site material placement (Giftthaler et al., 2017). A plausible implication is that the term spans at least two manufacturing logics: whole-volume liquid form-finding and end-to-end robotic construction. The shared emphasis is not a common mechanism but a common rejection of fragmented, pointwise fabrication workflows.

2. LiquiFab: whole-volume formation under weightlessness

LiquiFab shapes an entire volume of liquid polymer by subjecting it to a set of geometrical constraints under conditions of weightlessness, so that the physics of liquid interfaces drives the polymer to naturally adopt a configuration that minimizes its surface energy (Hochman et al., 19 Dec 2025). On Earth, weightlessness is achieved through neutral buoyancy; in space it is intrinsic. Under perfect neutral buoyancy, gravity and buoyancy cancel so that the only body-force on the liquid volume is surface tension. The free interface shape $S$ minimizes total surface energy subject to a fixed volume $V_0$ and contact with prescribed boundary surfaces $S_1, S_2,\dots,S_n$:

$$\text{Minimize}\qquad E[S] \;=\;\gamma\,\int_{S} dA$$

subject to

$$\int_{V} dV \;=\;V_{0}, \qquad \partial S\subset\bigcup_i S_i.$$

Here $\gamma$ is the liquid’s surface tension. Introducing a Lagrange multiplier $\lambda$ for volume conservation yields the Young–Laplace condition

$$\Delta P \;=\; \lambda \;=\; 2\,\gamma\,H \quad\Longrightarrow\quad H=\tfrac{\lambda}{2\gamma},$$

where $H$ is the mean curvature of the interface (Hochman et al., 19 Dec 2025). On each boundary surface $S_i$, the liquid meets at a pinned contact line, and the contact angle is fixed by wetting properties; for full wetting one takes effectively a zero contact angle.

The geometrical constraints are typically small disks, rings, or patches placed in space so that their collective geometry encodes the desired final form. For three equally spaced planar rings, the boundary surfaces may be written as

$V_0$0

By choosing the rings’ positions and radii $V_0$1, one sculpts a family of possible minimum-energy surfaces, then selects the injected volume $V_0$2 to pick the unique shape that matches the design intent (Hochman et al., 19 Dec 2025). This establishes Total Fabrication, in the LiquiFab sense, as a constrained variational shaping procedure rather than a path-planned deposition process.

3. Workflow and sequential assembly in LiquiFab

The LiquiFab workflow is explicitly organized into neutral buoyancy, liquid shaping, curing, and sequential assembly (Hochman et al., 19 Dec 2025). First, a transparent container is filled with an immersion liquid such as water–glycerol whose density exactly matches the liquid polymer. A small set of rigid boundary surfaces—rings, disks, or plates—is placed in the container, and these need only be a few percent of the object’s surface area. Second, the entire polymer volume $V_0$3 is injected via syringe through one or more boundary elements. Initially a spherical cap forms at each injection site; as injection continues, these blobs grow, wet additional boundaries, coalesce if needed, and instantly reconfigure into the unique minimum-energy interface spanning all constraints. Third, once the liquid stops evolving and assumes its target form, UV illumination is used for photopolymers or thermal cure for two-part silicones until the shell solidifies. The solid part is then released from its silicone or 3D-printed supports, washed free of immersion liquid, and dried (Hochman et al., 19 Dec 2025).

The sequential assembly stage is designated “LiquiBricks.” Each cured object exposes new free edges or surfaces that serve as boundary conditions for the next polymer injection. Using a robotic arm or manual repositioning, one visits each pair or set of adjacent boundaries, repeats the shaping and curing cycle with a fresh volume, and thus clicks bricks together. The core of each brick remains liquid until final global cure, ensuring perfect fusion of successive elements (Hochman et al., 19 Dec 2025).

This workflow has two distinct technical consequences. First, the boundary set may remain small relative to the total surface to be formed. Second, assembly is recursive: each cured piece becomes a new geometrical constraint for later pieces. This suggests that Total Fabrication, in this formulation, is not only a single-step shaping event but also a compositional strategy for building arbitrarily complex assemblies without ever visiting each point in space with a print head or laser (Hochman et al., 19 Dec 2025).

4. Scalability, speed, and limitations of whole-volume liquid fabrication

LiquiFab’s principal scaling claim is that because the entire volume forms in one step, the cycle time per element is dominated by injection and cure, measured in minutes, rather than by object size (Hochman et al., 19 Dec 2025). On Earth, the maximum scale is set by container size and by how precisely densities can be matched, described in terms of finite capillary length; in practice, objects of 20+ cm were demonstrated with sub-nanometric surface smoothness. In microgravity, no immersion liquid is needed at all, and objects can grow unbounded in any direction, limited only by the reach of the injection system (Hochman et al., 19 Dec 2025).

The method is contrasted directly with layer-based printing. Whereas layer-based printing has time scaling as $V_0$4, LiquiFab’s time per element is described as largely $V_0$5 with respect to volume, making it suitable for large, rapid fabrication of habitat segments, domes, or structural beams both on Earth and in space (Hochman et al., 19 Dec 2025). This is a concrete statement about asymptotic process behavior, not merely an assertion of practical speed.

The limitations are equally explicit. Geometric freedom is constrained to shapes realizable by a small set of bounding surfaces, and extremely concave or labyrinthine geometries may require many auxiliary supports. Transient fluid dynamics, including slow draining in thin shells, can freeze intermediate shapes if cured quickly; predicting these non-equilibrium forms demands a fully time-dependent fluid solver rather than the steady-state Surface Evolver approach. Current implementations use a single injector, yielding convergent or merging topologies; future work on dual or multiple injectors is identified as necessary for branching tree-like structures, diverging networks, and graded porosities. Injector design may be improved by adding free rotation on ball joints, smaller nozzles, or deployable boundary frames. Material development for deep-space environments, specifically vacuum- and radiation-resistant resins, is identified as a prerequisite for on-orbit structural fabrication (Hochman et al., 19 Dec 2025).

A common misconception would be to equate whole-volume formation with unconstrained freeform generation. The reported constraints indicate the opposite: the method depends on precisely placed boundary elements, pinned contact lines, wetting conditions, and controlled curing. Its novelty lies in the replacement of voxelwise toolpaths with global surface-tension control under weightlessness, not in the elimination of geometric constraints.

5. In situ Fabricator: end-to-end, full-scale digital construction on site

The In situ Fabricator presents Total Fabrication in an architectural robotics setting. The first prototype, IF1, is conceived as a self-contained, all-terrain mobile manipulator for true 1:1 on-site construction (Giftthaler et al., 2017). Its hardware includes a 6-axis ABB IRB 4600 serial manipulator with 2.55 m reach and 40 kg payload, a hydraulically driven tracked base with 5 km/h maximum speed powered by an AGNI DC-motor–pump pack at 150 bar system pressure and four hydraulic motors, Li-ion battery packs giving 3–4 h autonomy, an industrial-grade PC with Intel i3 @2.8 GHz and 4 GB RAM running Xenomai-patched Linux, and sensor suites tailored per task, including a scanning 2D laser rangefinder on the wrist, stereo or camera rigs for AprilTag-based localization, wire-mesh line detection and deformation feedback, and optional force–torque sensing at the flange (Giftthaler et al., 2017). A universal end-effector flange supplies power, hydraulics, and data for brick-laying grippers, concrete-extrusion nozzles, or a bending and welding head for Mesh Mould.

The software architecture integrates localization, motion planning, and control. A reference scan $V_0$6 of the site is acquired before building. During each reposition, a new scan $V_0$7 is registered against $V_0$8 by solving

$V_0$9

so that the robot knows its base pose in the CAD-model frame to within 1–5 mm (Giftthaler et al., 2017). For high-speed feedback and finer features, multiple AprilTags with known world-frame coordinates $S_1, S_2,\dots,S_n$0 are observed by base-mounted cameras, giving pose measurements

$S_1, S_2,\dots,S_n$1

where $S_1, S_2,\dots,S_n$2 is the sought robot-base pose, $S_1, S_2,\dots,S_n$3 and $S_1, S_2,\dots,S_n$4 are calibrations, and $S_1, S_2,\dots,S_n$5 is image noise (Giftthaler et al., 2017).

All-body motions are generated using Constrained Sequential Linear-Quadratic Optimal Control. The problem is cast as minimizing

$S_1, S_2,\dots,S_n$6

subject to

$S_1, S_2,\dots,S_n$7

with $S_1, S_2,\dots,S_n$8 comprising base-and-arm states and $S_1, S_2,\dots,S_n$9 the track velocities and joint torques or positions (Giftthaler et al., 2017). At each MPC iteration, up to 100 Hz, SLQ linearizes $$\text{Minimize}\qquad E[S] \;=\;\gamma\,\int_{S} dA$$0, solves the Riccati backward pass in $$\text{Minimize}\qquad E[S] \;=\;\gamma\,\int_{S} dA$$1, and yields a time-varying feedback gain $$\text{Minimize}\qquad E[S] \;=\;\gamma\,\int_{S} dA$$2 and feedforward $$\text{Minimize}\qquad E[S] \;=\;\gamma\,\int_{S} dA$$3; the real-time law $$\text{Minimize}\qquad E[S] \;=\;\gamma\,\int_{S} dA$$4 ensures millimeter-level tracking despite track–ground uncertainties. For sequential tasks, base motions are planned first via a constrained variant of STOMP, arm trajectories are optimized in the resulting local map, and a library of manipulation primitives is invoked under the SLQ feedback regulator (Giftthaler et al., 2017).

This architecture defines Total Fabrication here as a seamless CAD-to-fabrication chain rather than a material physics strategy. The fabrication process remains pointwise at the end effector, but the workflow from digital model to full-scale on-site realization is integrated end to end.

6. Architectural applications, quantitative performance, and open constraints

Two application scenarios are described for IF1. The first is a full-size undulating brick wall: a parametric CAD model of a 6.5 m $$\text{Minimize}\qquad E[S] \;=\;\gamma\,\int_{S} dA$$5 2 m double-leaf brick wall comprising 1,600 bricks is exported into the planning environment. IF1 captures a 3D scan of the site and registers the CAD anchor-pillars so that all brick coordinates are transformed accordingly. The repositioning loop occurs 14 times: the base pose $$\text{Minimize}\qquad E[S] \;=\;\gamma\,\int_{S} dA$$6 is planned so that the next reachable brick patch lies within the arm’s workspace, the platform drives to the pose and localizes by ICP or AprilTags to $$\text{Minimize}\qquad E[S] \;=\;\gamma\,\int_{S} dA$$7 mm, brick-placement primitives are executed, and placement is verified with the wrist-mounted laser rangefinder; if deviation exceeds 3 mm, a minor corrective patch is applied. Across 1,600 bricks and 14 base moves, the maximum cumulative placement error remained below 7 mm, and the dry-stack wall was completed in approximately 8 h of semi-autonomous operation, with human intervention only to reload bricks (Giftthaler et al., 2017).

The second scenario is the Mesh Mould concrete process. A bespoke steel-wire mesh is robotically built to serve as both mould and reinforcement, after which concrete is sprayed or pumped in to complete a doubly-curved wall. A custom bending and welding head is mounted to IF1’s flange; on-board hydraulics power the bending cylinders, while micro-controllers and valve drivers for weld current are networked into the real-time bus. A dual vision loop is used: base cameras track AprilTags on the pre-built starter mesh for 5 mm repositioning localization, while a wrist stereo rig detects the most recently welded wire segments via RANSAC line fitting to align the next bend and weld operation while compensating for elastic deformation. The build sequence consists of robot motion to a mesh zone, localization, wire threading, bending under SLQ feedback along a 3D spline path $$\text{Minimize}\qquad E[S] \;=\;\gamma\,\int_{S} dA$$8 while holding force $$\text{Minimize}\qquad E[S] \;=\;\gamma\,\int_{S} dA$$9, welding, stereo inspection, local spline recomputation if wire drift exceeds 2 mm, and repeated base shifts until the full mesh panel is completed. Early structural tests on small doubly curved panels demonstrated compliance within design tolerances below 10 mm (Giftthaler et al., 2017).

The principal reported metrics are summarized below.

Quantity	Reported value
Positioning accuracy (end effector)	1–5 mm
Max cumulative error in brick wall	7 mm over 1600 bricks
Base localization	$$\int_{V} dV \;=\;V_{0}, \qquad \partial S\subset\bigcup_i S_i.$$0 mm (ICP) or $$\int_{V} dV \;=\;V_{0}, \qquad \partial S\subset\bigcup_i S_i.$$1 mm (AprilTag)
Fabrication speed (brick wall)	~200 bricks/hour of autonomous placement
Motion control loop	100 Hz SLQ-MPC feedback
Planning updates	250 Hz (arm) and 100 Hz (base)
Autonomous runtime	3–4 h on battery alone
Top driving speed	5 km/h

These results establish the on-site robotic meaning of Total Fabrication as quantitatively grounded full-scale construction with millimeter-class localization and control (Giftthaler et al., 2017). At the same time, the paper emphasizes limitations of the IF1 platform: poor payload-to-weight ratio, position-only control with stiff gearboxes that preclude high-bandwidth force control, and lack of integrated sensing and hydraulic power at each joint. The proposed IF2 therefore calls for high power density of at least 1500 Nm continuous torque in at most 25 kg mass per joint, fully collocated force/position sensing with at least 1 kHz control bandwidth, on-board hydraulic power and electronics, and additively manufactured, load-optimized structural integration (Giftthaler et al., 2017). A preliminary sizing gives

$$\int_{V} dV \;=\;V_{0}, \qquad \partial S\subset\bigcup_i S_i.$$2

implying that each major joint must continuously deliver approximately 1.5 kNm at 0.5 rad/s while staying under 25 kg mass (Giftthaler et al., 2017).

A plausible implication is that the main controversy around Total Fabrication is not whether end-to-end digital construction is feasible, but which bottleneck dominates at scale: material formation physics in the LiquiFab case, or actuator, sensing, and force-control integration in the In situ Fabricator case. Both bodies of work present Total Fabrication as a systems problem, but they locate the critical system boundary in different places.

7. Synthesis and research directions

Across the cited literature, Total Fabrication is not a single standardized method but a unifying label applied to fabrication systems that seek to eliminate serial, voxelwise, or manually segmented production logic. In LiquiFab, the decisive operation is global shape adoption by a single liquid volume under surface-energy minimization, followed by curing and recursive assembly into larger structures (Hochman et al., 19 Dec 2025). In the In situ Fabricator, the decisive operation is a seamless CAD-to-site workflow supported by mobile manipulation, exteroceptive localization, optimal control, and task-specific end effectors for full-scale construction (Giftthaler et al., 2017).

The research trajectories likewise differ. For LiquiFab, future directions include dual or multiple injectors for branching and diverging topologies, improved injector mechanics such as ball-joint rotation and deployable boundary frames, fully time-dependent fluid solvers for non-equilibrium shape prediction, and vacuum- and radiation-resistant resins for deep-space use (Hochman et al., 19 Dec 2025). For the In situ Fabricator, future work centers on actuator redesign through fully integrated titanium vane actuators, hybrid wheels-and-legs locomotion for narrow-doorway and stair traversal, passive-safety modes on human-contact detection, and automatic power allocation between base and manipulators (Giftthaler et al., 2017).

This suggests a broader interpretation of Total Fabrication as a research program aimed at replacing local toolpath dominance with globally coordinated fabrication processes. In one branch, global coordination is achieved by liquid interface physics and boundary conditions; in the other, by high-performance sensing, planning, and control. The term therefore marks a shift in where fabrication intelligence resides: either in the energy-minimizing behavior of matter under prescribed constraints or in the tightly integrated robotic system that links design models to material placement at full scale.