Rapid Deployment of Curved Surfaces via Programmable Auxetics

MINA KONAKOVIĆ-LUKOVIĆ, EPFL

JULIAN PANETTA, EPFL
KEENAN CRANE, Carnegie Mellon University
MARK PAULY, EPFL

Fig. 1. Our algorithm computes a triangular auxetic linkage that closely approximates a given surface when deployed to maximal extension via inflation. The
fabricated material is laser cut from a single sheet, mounted onto the support frame, and inflated with a generic rubber balloon.

Deployable structures are physical mechanisms that can easily transition CCS Concepts: • Computing methodologies → Mesh geometry models; •
between two or more geometric configurations; such structures enable in- Applied computing → Computer-aided design; Computer-aided manufac-
dustrial, scientific, and consumer applications at a wide variety of scales. turing;
This paper develops novel deployable structures that can approximate a
large class of doubly-curved surfaces and are easily actuated from a flat Additional Key Words and Phrases: computational fabrication, smart materi-
initial state via inflation or gravitational loading. The structures are based als, digital fabrication, auxetic materials, conformal geometry
on two-dimensional rigid mechanical linkages that implicitly encode the
ACM Reference Format:
curvature of the target shape via a user-programmable pattern that permits
Mina Konaković-Luković, Julian Panetta, Keenan Crane, and Mark Pauly.
locally isotropic scaling under load. We explicitly characterize the shapes
2018. Rapid Deployment of Curved Surfaces via Programmable Auxetics.
that can be realized by such structures—in particular, we show that they can
ACM Trans. Graph. 37, 4, Article 106 (August 2018), 13 pages. https://doi.org/
approximate target surfaces of positive mean curvature and bounded scale
10.1145/3197517.3201373
distortion relative to a given reference domain. Based on this observation,
we develop efficient computational design algorithms for approximating a
given input geometry. The resulting designs can be rapidly manufactured 1 INTRODUCTION
via digital fabrication technologies such as laser cutting, CNC milling, or 3D Deployable structures are shape-shifting mechanisms that can tran-
printing. We validate our approach through a series of physical prototypes
sition between two or more geometric configurations. Often con-
and present several application case studies, ranging from surgical implants
ceived to minimize space requirements for storage or transport,
to large-scale deployable architecture.
deployable structures are used, for example, for antennas or solar
Authors’ addresses: Mina Konaković-Luković, [email protected], EPFL, Route panels in satellites, as coronary stents in medical applications, as
Cantonale, Lausanne, Switzerland, 1015; Julian Panetta, EPFL, [email protected];
Keenan Crane, Carnegie Mellon University, 5000 Forbes Avenue, Pittsburgh, Pennsyl- consumer products (e.g. umbrellas), or in architectural designs (e.g.
vania, 15213, [email protected]; Mark Pauly, EPFL, [email protected]. retractable bridges or relocatable, temporary event spaces).
Most existing realizations of deployable structures are geometri-
Permission to make digital or hard copies of all or part of this work for personal or
classroom use is granted without fee provided that copies are not made or distributed cally simple and often exhibit strong symmetries. Deploying more
for profit or commercial advantage and that copies bear this notice and the full citation general curved surfaces is made difficult by the inherent complex-
on the first page. Copyrights for components of this work owned by others than ACM
must be honored. Abstracting with credit is permitted. To copy otherwise, or republish,
ity of jointly designing initial and target geometries within the
to post on servers or to redistribute to lists, requires prior specific permission and/or a constraints imposed by the deployment mechanism [Gantes 2001].
fee. Request permissions from [email protected]. We propose a new class of deployable structures and associated
© 2018 Association for Computing Machinery.
0730-0301/2018/8-ART106 $15.00
computational methods that enable rapid deployment of doubly-
https://doi.org/10.1145/3197517.3201373 curved freeform surfaces (see Figure 1). Our approach is based on a

ACM Trans. Graph., Vol. 37, No. 4, Article 106. Publication date: August 2018.
106:2 • Konaković-Luković et. al.

planar linkage of rigid triangles connected by rotational joints at

vertices; this linkage has regular connectivity, but spatially varying
scale. In-plane rotation of the triangles induces an approximately
isotropic expansion or contraction in area, which allows a mechan-
ical interpretation of the linkage as an auxetic surface metamate-
rial [Saxena et al. 2016], or a geometric interpretation in terms of
conformal maps [Konaković et al. 2016]. By spatially varying the tri-
angle sizes, we effectively control the maximum possible expansion
at each point, which in turn provides control over curvature: under
maximal extension, nonuniform expansion forces the structure to
buckle out of the plane and assume a curved configuration. Here
several questions arise: which curvature functions can be encoded
in such a pattern? How can we actuate a linkage to achieve maximal Fig. 2. Conventional heart stents are straight and typically chosen by the
expansion? Which surfaces can we hope to realize using this proce- surgeon from a set of standard sizes. Recent research has shown the benefits
of curved stents [Tomita et al. 2016]. Our method can be used to create
dure? Several key contributions help to address these questions:
freeform curved heart stents that can be adapted to the specific geometry
of the patients’ blood vessels. The stent is administered with a catheter to
• We introduce spatially graded auxetic metamaterials suitable
the correct position (left) inflated to its target geometry (middle, right).
for deployment via inflation or gravitational loading. In par-
ticular, we show that these deployment strategies achieve
maximal expansion everywhere and provide additional regu-
larization to ensure that the target shape is unique. air-supported domes. We conclude with a discussion of the limi-
• We provide a general analysis of deformation by inflation and tations of our approach, and also identify opportunities for future
gravitational loading to formally classify the set of realizable research (7).
doubly-curved target shapes.
• We present an optimization algorithm to solve the inverse
2 RELATED WORK
design problem: Given a desired target geometry, our method Computational material design. Several previous works have de-
finds appropriate scaling parameters and a corresponding signed custom materials to achieve high-level deformation goals.
layout of the 2D linkage such that the target shape is achieved Bickel et al. [2010] stack layers of various nonlinear base materials to
when the linkage is deployed. produce a desired force-displacement curve. Microstructure design
works [Panetta et al. 2015; Schumacher et al. 2015; Zhu et al. 2017]
The resulting structures offer a number of benefits: (i) The rest- construct small-scale structures from one or two printing materials
state is (piece-wise) flat, which facilitates compact storage as well as to emulate a large space of linearly elastic materials. These works fo-
cost- and time-efficient fabrication techniques (such as laser cutting cus on designing deformable materials that typically undergo small
or milling); (ii) The target geometry is directly encoded in the 2D stretches and return to their rest configurations when unloaded,
linkage structure so that no additional support or scaffolding is making them less suitable as deployment mechanisms.
required to guide deployment; (iii) Our approach is scale-invariant
and can be applied to realize a broad and explicitly defined class Inverse elastic shape design. Another common goal is to opti-
of doubly-curved surfaces. If a given surface is not within the set mize deformable objects’ rest shapes so that they assume desired
of realizable shapes, we apply optimization to find a feasible target equilibrium shapes under load. The inverse elastic shape design
surface that is close to the desired design. algorithms of [Chen et al. 2014; Pérez et al. 2015] design flexible ob-
The rest of the paper is organized as follows: Section 2 reviews jects achieving specified poses under gravity or user-defined forces.
connections to related work. Section 3 considers geometric models These works do not attempt to find compact rest configurations
of inflation and gravitational loading, helping to understand the amenable to efficient fabrication, transport, and deployment. One
feasible design space. Specifically, we characterize the shapes that exception is [Skouras et al. 2012], which designs rubber balloons
can be achieved via inflation or gravitational loading in terms of that inflate to desired target shapes. However, fabricating custom
surfaces of positive mean curvature and conformal deformations rubber balloons involves a complicated multi-step molding process
with bounded scale factor. Section 4 introduces our spatially graded best suited for small-scale target shapes. Additionally, the inflation
auxetic metamaterial, realized as a rigid triangular linkage. We show must be carefully controlled to avoid under- or over-inflating. In
how to locally adapt maximal expansion (and hence, target curva- contrast, our flat initial state facilitates simple fabrication at a wide
ture) by varying the scale and orientation of linkage elements in range of scales. Our deployment method is also more robust, since
the initial flat state. Section 5 describes an optimization algorithm the final state is precisely singled out by construction – the target
for solving the inverse design problem, i.e., finding suitable parame- is reached when the material cannot expand any further.
ters for our metamaterial that ensure the target surface is faithfully Deployment-aware design. Other works have focused on design-
approximated when actuated. In Section 6 we present several case ing objects that rapidly expand into nearly rigid target shapes. Sk-
studies and physical prototypes that highlight potential applications ouras et al. [2014] construct inflatable structures by fusing together
across domains ranging from small-scale heart stents to large-scale sheets of nearly inextensible material. Because each panel inflates

ACM Trans. Graph., Vol. 37, No. 4, Article 106. Publication date: August 2018.
 Rapid Deployment of Curved Surfaces via Programmable Auxetics • 106:3

results, but has several drawbacks. (i) Pre-stretched materials are

limited in scale. (ii) Fabrication is more complex, since it requires
compositing multiple materials. (iii) Shaping by contraction means
that the flat surface is larger in area than the target surface, reducing
potential packing benefits. (iv) Closed surfaces are more difficult to
realize (only disk-topology surfaces have been shown).
Our approach is scale-invariant, does not require multi-material
compositing (our inflation balloons need not be attached to the
auxetic linkage), leads to compact flat-packed sheets, and can handle
shapes of arbitrary topology.

Auxetic Materials. Auxetic surface materials are an essential in-

gredient of our approach: auxetic linkages permit otherwise inex-
tensible flat sheets of material to uniformly stretch as needed to
deform into doubly curved surfaces. We refer the reader to [Saxena
et al. 2016] for a survey on auxetic patterns, their unique mechanical
properties, and their potential applications to diverse engineering
Fig. 3. Design study of deployable architecture. The freeform inflatable and medical problems. In graphics, Konaković et al. [2016] intro-
dome can be used as a semi-permanent, relocatable space.
duced a design tool for fabricating curved target surfaces by cutting
auxetic patterns into flat sheets. However, the resulting uniform
into a nearly developable surface, many small panels are poten- linkage pattern is difficult to deploy because the target surface is
tially needed to closely approximate a smooth, wrinkle-free doubly- not singled out in any way; the structure can just as easily deform
curved surface. Zheng et al. [2016] design compact scissor linkage into an infinite family of other surfaces. This ambiguity necessitates
assemblies that, when stretched, uniformly expand into coarse ap- the use of guide surfaces and careful manual alignment when shap-
proximations of 3D shapes. Their method ensures a collision-free ing the material. Our work addresses this limitation by spatially
expansion path for sparse wireframe designs. Dudte et al. [2016] per- varying the pattern to uniquely encode the target shape, enabling
form basic research into approximating singly- and doubly-curved rapid deployment without guide surfaces by simple expansion.
surfaces with generalized Miura folds. Their origami patterns have Friedrich et al. [2018] also seek to encode the target surface by
a single degree of freedom parametrizing their path from the flat limiting the pointwise maximal stretch factors. Rather than design-
configuration to the target shape. For doubly curved surfaces, the ing a fully opened linkage on the target surface, the authors outline
construction is bi-stable, leading to an especially simple deployment a heuristic to construct a partially opened pattern in the plane:
process. However, the design algorithm produces flat configurations based on the scale factors of a conformal map, they insert polygonal
with over twice the surface area of the target. openings resembling those at the top of Figure 6. However, it is un-
clear how these polygonal openings are positioned and connected
Actuated shape-shifting. The engineering and graphics communi- to ensure proper linkage functionality. The authors then propose
ties have both sought to design mechanisms that transition between an iterative evolutionary optimization process needed to bring the
discrete configurations or trace out continuous deformation paths fully opened pattern closer to the target surface. In contrast, our
when actuated. By embedding a rigid fiber lattice in flexible silicon, algorithm directly ensures the target surface is closely approximated
Connolly et al. [2017] design tubes that accurately reproduce bend- by the fully opened linkage; furthermore, we detail efficient mech-
ing and twisting motions when inflated. Ma et al. [2017] generalize anisms for deployment and characterize the space of achievable
this idea, segmenting objects into chambers that, when inflated to designs for each deployment mechanism.
certain pressures, drive the shape into a sequence of desired poses. Multi-stable auxetic patterns [Rafsanjani and Pasini 2016] are
Also using pneumatic actuation, Overvelde et al. [2016] present an another potential avenue for encoding maximal stretch factors in a
origami-inspired metamaterial that dramatically changes shape, and material: they expand from their rest configuration and settle into
Ou et al. [2016] design flat sheets that fold into complex origami stable equilibrium at one or more stretched configurations. We leave
shapes. Raviv et al. [2014] design structures that can bend, stretch, investigation on how to modify these patterns to encode curved
and fold when exposed to water. Liu et al. [2017] study how a surfaces as future work.
pre-strained elastomer sheet patterned with ink can self-fold when
heated by a lamp. Polyhedral Patterns. When deployed, our auxetic linkage’s equi-
Actuated form-prescribed geometry. Recent work [Guseinov et al. lateral triangles and hexagonal openings tile the target surface with
2017; Pérez et al. 2017] follows a similar rationale of encoding a a tri-hex pattern. A common task in architectural geometry is to
3D target surface in a flat sheet of material. In these methods, the rationalize curved input surfaces using planar polyhedral patterns.
activation mechanism is directly integrated into the material in Schiftner et al. [2009] and Jiang et al. [2015] both introduce algo-
the form of a pre-tensioned elastic membrane. Upon release, the rithms that can approximate input surfaces with tri-hex patterns
membrane contracts and forces the pre-shaped rigid elements into as a special case. Vaxman et al. [2017] propose a form-finding tool
their global target configuration. This approach achieves impressive for general combinatorial patterns and show applications to tri-hex

ACM Trans. Graph., Vol. 37, No. 4, Article 106. Publication date: August 2018.
106:4 • Konaković-Luković et. al.

meshes. These works focus on symmetry, planarity, and other pat-

tern quality requirements, making no attempt to ensure the tri-hex
pattern can be flattened into a planar configuration by closing the
hexagons, which is essential in our approach.

3 SHAPE SPACE
Which shapes can we hope to achieve with our structures? The
answer depends jointly on the geometry of the structure, as well
as the method used to actuate it. Rather than study this question
in terms of the detailed geometry of a specific mechanical linkage,
we will first consider an idealized model based on smooth differen-
tial geometry. This analysis will then inform the design of discrete
mechanical linkages and their physical actuation described in Sec-
tion 4 and the corresponding optimization algorithm discussed in
Section 5. In particular, we will explicitly characterize the shapes
one can hope to achieve via (i) inflation and (ii) gravitational load-
ing; we will also make an interesting connection between inflated Fig. 4. Design study of a freeform chair realized using four layers of spatially
balloons and conformal geometry (Section 3.2.1). graded auxetic material to fully cover the surface (see Section 5 for details).

3.1 Preliminaries
geometry) is beyond the scope of this paper. Moreover, for compu-
In this section, we consider a closed, compact, and oriented topologi-
tational design, it is often more useful to have a simple and easily
cal surface M with geometry given by a map f : M → R3 assigning
computable geometric model than a detailed mechanical model
coordinates to each point of M. The differential d f of f maps tan-
which is accurate but difficult to explore due to heavy computa-
gent vectors X on M to the corresponding vectors d f (X ) in R3 ; the
tional requirements (e.g., finite element analysis).
differential is also sometimes denoted as the Jacobian or deformation
We specifically consider the geometry of immersions that (i) lo-
gradient. A map f is an immersion if its differential is injective, i.e.,
cally maximize enclosed volume, and (ii) do not stretch area above
if at each point p ∈ M it maps nonzero vectors to nonzero vectors;
a given upper bound. Note that we do not consider questions of dy-
since M is compact, it is an embedding if f is also injective (loosely
namics (e.g., “can this configuration be reached from a given starting
speaking: if it has no self-intersections). Formally, we will require
point?”), which are notoriously difficult even without constraints on
that f is a twice differentiable immersion with bounded curvature.
volume or area. Instead we consider only the simple static question
To any immersed surface we can associate the quantity
∫ of, “what will be true about a surface that achieves these conditions?”
In particular, we make the following observation:
vol(f ) := N · f dAf ,
M Proposition 3.1. Let dA+ be an area measure on M. Among all
where N is the outward unit normal, and dAf is the area element (twice differentiable) immersions f : M → R3 such that dAf ≤ dA+ ,
induced by f ; when f is embedded, vol(f ) is just the enclosed those that locally maximize the enclosed volume vol(f ) will (i) have
volume. We will also use д and H to denote the metric and mean strictly positive mean curvature H > 0 away from sets of measure
curvature (resp.) induced by f . We use the definition H = 21 ∇f · N , zero, where H ≥ 0; and (ii) will achieve the upper bound on area
so, e.g., a sphere has constant positive mean curvature. If dA and (dAf = dA+ ).
d à are two area measures on M, we will write dA ≤ d à to mean
Proof. (i) Suppose an immersion f admits a nonempty open set
that dA(U ) ≤ d Ã(U ) for all open sets U ⊂ M. When considering
D ⊆ M on which H ≤ 0. Then we can construct a smooth positive
variations of the surface, we will think of f as a time-parameterized
d ϕ| function u : M → R supported on D and consider the outward
family of immersions f (t) and adopt the shorthand ϕÛ := dt t =0 normal variation fÛ := uN . The corresponding first-order changes
for any time-varying quantity ϕ. in volume and area measure are given by
∫
3.2 Inflation d vol(f )|
t =0 = udAf > 0 and
dt
M
To understand the space of shapes that can be achieved via inflation,
d
we consider an idealized and purely geometric model of rubber bal- dt dAf |t =0 = 2uHdAf ,
loons. From a mechanical viewpoint, our model would correspond respectively. Since uH ≤ 0, this variation increases volume without
(very roughly) to a thin isotropic elastic membrane with spatially increasing area; hence, f is not a volume maximizing immersion.
varying maximal expansion. This model should however be taken Moreover, if H < 0 at any point p ∈ M, then (by continuity of H )
with a grain of salt: our goal here is not to formulate a precise me- there must be an open ball around p on which H is strictly negative.
chanical model, but rather to get a sense of the most significant Hence, on sets of measure zero, an immersion f that maximizes
geometric effects exhibited by our discrete mechanism—a more rig- volume must have H ≥ 0.
orous analysis (e.g., based on homogenization of the small-scale

ACM Trans. Graph., Vol. 37, No. 4, Article 106. Publication date: August 2018.
 Rapid Deployment of Curved Surfaces via Programmable Auxetics • 106:5

area scaling of at most 4). However, Dirichlet energy will still tend
to dominate more realistic nonlinear models of elasticity in the
limit of large stretching. Consider for instance the elastic energy
described by Chao et al. [2010], E(f ) := M |d f − R f | 2 dA, where
∫

R f denotes the rotation closest to d f . In the limit of large strain, the

quadratic (i.e., Dirichlet) term of the expansion |d f | 2 − 2⟨d f , R f ⟩ +
|R f | 2 dominates, and we are left with the same picture as before.
Moreover, the nonlinear terms drop off rather quickly, suggesting
that one should observe conformal behavior even for moderate
stretching—as supported by physical experiments such as the one
Fig. 5. When inflated, rubber balloons exhibit near-conformal deformation pictured in Figure 5. This observation further motivates our use of
(indicated by the preservation of right angles), further motivating our use
auxetic materials with bounded scale factors for inflatable structures.
of an auxetic design space for inflatable structures.

3.3 Gravitational Deployment

(ii) Since both dAf and dA+ are area measures, we have dA+ = Gravity is an even simpler mechanism for shape deployment: just
φdAf for some continuous function φ : M → R. If dAf < dA+ , suspend a sheet of material by its boundary and let gravity pull it
then there will be at least one point p ∈ M where φ(p) < 1, and into the target shape. This approach is most suitable for surfaces
by continuity, an open neighborhood D around p where φ < 1. with simple boundary curves like in Figure 11. In fact, to simplify
Letting u be a smooth positive function supported on D, a normal the fabrication process, we require that the initial surface spanning
variation uN will now increase the volume without violating the the boundary curves be a height field; otherwise attaching the flat
area bound. □ material to the boundary curves would require a complicated manual
Roughly speaking, the surfaces that can be realized via infla- deformation. The height field property also guarantees that the
tion in our model are those that have positive mean curvature (see downward gravitational force has a positive component along the
Proposition 3.1 and Section 4.1 for further discussion). In practice, surface’s normal direction, ensuring that it can pull the surface open
we therefore modify a given target surface to have positive mean analogously to the inflation setup.
curvature, as described in Section 3.4. When fabricated from our idealized material (characterized by
having zero stiffness until an upper area bound is reached), we ob-
3.2.1 Conformal Balloons. serve that height-field-initialized surfaces will remain height fields
Suppose we no longer consider volume maximization nor an upper during the deployment. This follows from the fact that only two
bound on area, but simply ask about the shape of a balloon that types of forces act on interior points during the deployment: gravity
tries to minimize the elastic membrane energy when filled with and the material stresses enforcing the area bound. Gravity pushes
a∫ fixed volume of air. In particular, the Dirichlet energy E D (f ) := points in the material straight downward, decreasing height values
M
|d f | 2 dA0 models an elastic membrane with zero rest length, or, but preserving the height field property. Stresses enforcing the area
asymptotically, the energy due to extreme stretching. Critical points stretch bound always take the form of tensile forces: regions of
of Dirichlet energy are called harmonic maps, and any harmonic material that have reached their stretching bound pull uniformly
map between topological spheres is necessarily holomorphic or anti- inward against the surrounding material (tangentially to the sur-
holomorphic [Eells and Wood 1976]. We hence find a connection to face). Unlike expansive forces, these tensile forces act to straighten
conformal geometry: out the material and will not cause the sheet to fold over itself to
violate the height field property.
Theorem 3.2. Consider a surface (M, д0 ) of spherical topology.
We now characterize the space of height field surfaces deployable
Among all embeddings f : M → R3 of fixed enclosed volume vol(f ) =
by gravity, again ignoring questions of dynamics. For consistency
c, any embedding minimizing the membrane energy E D is conformal,
with the inflation setup, we orient the surface’s height axis verti-
i.e., the induced metric д := d f ⊗ d f is conformally equivalent to д0 .
cally (parallel to gravity) and choose the surface orientation so that
Proof. Let f˜ be a non-conformal embedding with volume c, and normals point downward. The gravitational deployment process is
let Σ := f˜ (M) ⊂ R3 denote the image of f˜ . We know that a map formulated as minimizing the immersion f ’s gravitational potential
minimizing Dirichlet energy over all orientation-preserving embed- energy:
dings f : (M, д0 ) → Σ is holomorphic (and in particular, conformal ∫
since it is injective and orientation-preserving). Therefore, because U (f ) := f · z dA,
f˜ is not conformal, we can find an embedding f ∗ mapping to Σ with M
smaller Dirichlet energy. This f ∗ still has volume c, but lower E D ,
where z is the height axis vector oriented opposite gravity and scaled
so f˜ is not minimal. □
by the gravitational acceleration constant. Note that dA is the area
Due to the rather simplistic model of membrane energy, one might element induced by an isometric immersion of M (for which we
wonder whether this theorem provides any useful information about assume the material density is 1) and is independent of the particular
real physical balloons or our auxetic mechanisms (which allow an immersion f .

ACM Trans. Graph., Vol. 37, No. 4, Article 106. Publication date: August 2018.
106:6 • Konaković-Luković et. al.

Proposition 3.3. A height field surface represented as a smooth We observe that u = 0 on ∂Ri since the perturbed surface must still
immersion f : M → R3 that locally minimizes the gravitational po- fill the same boundary curve.
tential energy U (f ) over all smooth immersions satisfying dAf ≤ dA+ The corresponding first-order change in mean curvature is [Doğan
and Dirichlet conditions f = f tgt on ∂M, must (i) have strictly positive and Nochetto 2012]:
mean curvature H > 0 away from sets of measure zero, where H ≥ 0;
2HÛ = − △f u − (κ 12 + κ 22 )u + 2 fÛ · ∇f H = − △f u − 2|K |u,
and (ii) achieve the upper bound on area (dAf = dA+ ).
(1)
where κ 1 = −κ 2 are the minimal surface patch’s principal curvatures.
Proof. (i) Suppose there exists a region D ⊆ M of nonzero mea- The term involving fÛ vanishes because H ≡ 0, and we applied the
sure on which H ≤ 0. We can construct a smooth, positive bump simplification κ 12 + κ 22 = 2|κ 1κ 2 | = 2|K |. Preserving non-negative
function u compactly supported on D so that the positive normal mean curvature requires:
variation fÛ := uN decreases gravitational potential to first order:
∫ HÛ ≥ 0 =⇒ △f u + 2|K |u ≤ 0.
d U (f + tuN )| = uN · z dA < 0,
dt t =0 For small |K | (mildly curved repaired patches), we expect the Lapla-
D
cian term to dominate and force the normal velocity to achieve its
because N · z < 0 by the height field property. Furthermore, this minimum on the boundary ∂Ri (superharmonic functions obey a
variation does not violate the upper bound on area: the area measure minimum principle). But u = 0 there, forcing u ≥ 0 inside Ri .
changes by Furthermore, in our experiments, closest points on the original
d AÛ f = 2uHdAf ≤ 0. surface always lie to the negative side of the repaired patch in that,
∀p ∈ f (Ri ) and nearest original points p ∗ = argminp̃ ∈f0 (Ri ) ∥p̃ − p ∥,
Therefore, f does not locally minimize gravitational potential en-
we have N · (p ∗ − p) ≤ 0. This should be expected for moderate edits,
ergy. (The proof of part (ii) is analogous to Proposition 3.1.)
as the curvature flow process converging to the minimal surface
□
moves points only in the positive normal direction. In these cases,
3.4 Projection to Feasible Surfaces
moving any point on our repaired surface closer to the original
surface requires a motion in the negative normal direction which,
If the surface violates the positive mean curvature requirement, we for small |K |, violates the non-negative mean curvature constraint.
must modify it for compatibility with our deployment mechanisms.
However, we wish to keep the design as similar to the input surface 4 MATERIAL DESIGN
as possible. Accordingly, we change the surface only where needed,
Our goal is to design a mechanism that deforms from an initial flat
leaving the regions of positive mean curvature untouched. In the
configuration into a doubly-curved target surface when actuated by
regions violating the requirement, we make the smallest change
inflation or gravity. Konaković et al. [2016] study a similar problem,
necessary in mean curvature space.
where linkages based on the regular 2D Kagome lattice are deformed
We propose the following repair process to achieve these goals:
into general curved target shapes. This approach has two key ob-
apply mean curvature flow fÛ = −H N to each region of negative
stacles to overcome when it comes to rapidly deployable structures,
mean curvature, terminating when mean curvature reaches zero.
namely (i) a perfectly regular lattice encodes no information about
Then, to ensure H ≥ ε > 0, an arbitrarily small, smooth normal
the target shape, necessitating some kind of “scaffolding” such as a
variation can be applied, computed, e.g., by solving Equation 1 with
3D print to guide assembly, and (ii) there is no clear way to actuate
HÛ = 1 and zero Dirichlet boundary conditions.
such a surface, which must be laboriously pointwise-aligned to the
Our repair process indeed produces the closest admissible surface
mold and deformed by hand. These observations motivate us to (i)
in the sense of minimizing pointwise curvature distance |H − H 0 |
encode the target shape into the linkage by considering a spatially
almost everywhere in M (where H 0 is the mean curvature of the ini-
varying pattern rather than a regular one, and (ii) consider geome-
tial immersion): it preserves mean curvature in the positive regions
tries that can be rapidly deployed via inflation or gravity, as studied
and minimally adjusts each non-positive value. Curvature-based
in Section 3.
distance metrics like this are often considered good models of per-
ceptual distance [Kim et al. 2002]. However, for the examples we Discrete Conformal Geometry. A key motivation for starting with
tried, we can make an additional observation: the repair process the Kagome lattice is that, as observed by Konaković et al. [2016],
also locally minimizes pointwise distances to the original surface. deformations of this lattice behave at the large scale like conformal
We formalize the repair process as follows. For a smooth initial mappings with bounded scale factor. This loose analogy is made a
immersion f 0 : M → R3 , the regions Ri ⊂ M on which H < 0 bit more precise by making a connection to the Cauchy-Riemann
are always bounded by well-defined curves ∂Ri . The repair pro- equations: for both conformal maps and the lattice, infinitesimal
cess cuts away each f 0 (Ri ) and replaces it with a minimal surface planar motions are determined by real degrees of freedom at the
f (Ri ) spanning the same immersed boundary curve. This viewpoint boundary. Another connection recently made by Lam [2017] is that
corresponds to the limit ε → 0. infinitesimal rotations of the lattice can be described as discrete
First, we consider the space of admissible variations one might harmonic functions (in the usual sense of the cotangent Laplacian),
apply to the repaired surface when attempting to move it closer mirroring the fact that for the logarithmic derivative log(z ′ ) = u +ıθ
to the original. We consider an arbitrary suitably regular variation of a holomorphic map z, the two components u, θ describing scaling
fÛ := Ri → R3 and define normal velocity u := fÛ ·N for convenience. and rotation (resp.) are conjugate harmonic. To date, however, there

ACM Trans. Graph., Vol. 37, No. 4, Article 106. Publication date: August 2018.
 Rapid Deployment of Curved Surfaces via Programmable Auxetics • 106:7

initial 2D state initial 2D state

initial 2D state

fully closed variable, partial expansion

fully expanded
variable, partial expansion fully expanded
Fig. 6. Spatially variable maximal expansion of the linkage can be achieved
by scaling and rotating the linkage triangles in the initial 2D state. When
already fully opened (left), no more expansion is possible. When fully closed
(right), the linkage can expand to increase by a factor two in length (or a fac-
tor of four in area). Partially opening the initial configuration allows varying
the scale factor, indicated by the size of the orange triangles connecting the
barycenters of the openings.

target 3D state target 3D state

is still no complete discrete theory of conformal maps based on requires guide surface automatically deployed
the Kagome lattice that includes finite deformations, nor confor-
mal immersions in R3 . Nonetheless, adopting the conformal point Fig. 7. The method of Konaković et al. [2016] (left) uses a uniform, fully
closed initial 2D state and achieves its target state with variable partial
of view allows us to leverage well-developed tools from computa-
openings. Proper deployment thus requires a guide surface and precise
tional conformal geometry for the purpose of designing deployable
manual alignment. In contrast, our spatially varying initial openings in
mechanisms. the 2D state allow encoding the target surface in the flat configuration,
Mechanical Properties. From a mechanical point of view, linkages facilitating automatic deployment by maximal expansion without the need
of any guide surface (right).
based on the Kagome lattice are flexible enough to produce a wide
variety of curved surfaces and already have a locking mechanism
built-in: stretching the material to four times its original area fully
opens the linkage, blocking further expansion. In fact, one can easily
we first consider the uniqueness of the deployed configuration, be-
show that the linkage is rigid (albeit unstable) in its fully open con-
fore detailing how to program the desired maximal expansion factor
figuration; additional forces such as gravity or air pressure help to
into our discrete triangular linkage.
stabilize the fully open state. We take advantage of these mechanical
properties to aid deployment. In particular, we adapt the pattern to
achieve a spatially varying (rather than constant) maximum bound 4.1 Uniqueness
on expansion across the surface. When deployed, the varying expan- The spatially varying maximal extension factor uniquely determines
sion leads to out-of-plane buckling; thus the linkage must assume a the fully expanded linkage’s metric. In other words, the deployed
curved configuration. shape is completely determined up to isometric deformation. Does
The geometric and mechanical pictures can of course be linked: this mean that the metamaterial uniquely encodes the target shape?
the bound on expansion in the discrete linkage can be modeled by In general, the answer is no. For instance, the material alone cannot
a bound on the conformal scale factor e u of a smooth conformal distinguish between “bumps” with negative or positive mean cur-
map, and the buckling exhibited by the deployed linkage is approx- vature since both produce the same metric distortion. However, in
imately determined by the Yamabe equation ∆u = e 2u K relating this case our specific deployment methods provide additional regu-
the logarithm of the scale factor to the Gaussian curvature K of larization: they always produce surfaces of positive mean curvature,
a smooth surface approximating the target geometry. To explore eliminating this ambiguity.
designs for our mechanical linkage, we therefore adopt a strategy Convex surfaces are known to be unique up to global rigid trans-
based on geometry: first, we compute a conformal map from the formations. Surprisingly, the question of whether smooth closed
plane to the target surface, and read off the scale factors λ tgt := e u . surfaces can be flexible in R3 (i.e., admit infinitesimal deformations
We then use these factors to design or “program” a spatially-graded preserving the metric) remains an open problem in differential ge-
pattern that approximately matches the corresponding maximum ometry [Ghomi 2017]. So far, no examples have been found, and in
expansion at each point. When fully expanded, a mechanism based practice, all of our examples deployed to their proper target config-
on this pattern should approximate the desired target shape. Below urations.

ACM Trans. Graph., Vol. 37, No. 4, Article 106. Publication date: August 2018.
106:8 • Konaković-Luković et. al.

S M3D 3D linkage 3D linkage L3D

initialization optimization

conformal conformal target edge

f f -1 lengths

regular 2D linkage 2D linkage

meshing initialization optimization
M2D L2D

Fig. 8. Sketch of the optimization algorithm for computing the spatially graded auxetic linkage for a given input surface S .

4.2 Auxetic Linkages with Locally-Controlled Stretching iteration number. We run this flow until convergence updating only
We now consider how to adapt the regular triangle auxetic link- the positions of vertices with non-positive mean curvature.
age structure to impose a spatially-varying upper scaling bound Given the corrected input surface S, our goal now is to find the
tailored to the conformal scale factor λ tgt . We begin with the follow- 2D layout of the triangular linkage that, when deployed to maximal
ing observation: taking the standard linkage pattern (with length expansion, approximates S as closely as possible. Figure 8 illustrates
stretch factor λ in the range 1 ≤ λ ≤ 2) and pre-stretching by 2/λ tgt the main steps of our algorithm.
yields a new material with the stretching bounds λ tgt /2 ≤ λ ≤ λ tgt .
Effectively, this pre-stretching limits the amount of additional ex- Conformal Flattening and Remeshing. We first compute a confor-
pansion possible until the fully opened configuration is reached (see mal map f : S → Ω from the target surface S to a planar domain
Figure 6). This reduces our problem to producing a linkage with a Ω ⊂ R2 using the methods of [Sawhney and Crane 2017]. We check
spatially-varying pre-stretch in its flat configuration. The challenge if the conformal scale factors are within the bounds prescribed by
now is to piece together patches with different pre-stretch. As illus- the linkage mechanism, and, if necessary, introduce cone singulari-
trated in Figure 6, this can only be done by scaling the triangles, as ties at user-selected locations to reduce scale distortion as described
will be detailed below. below. Next, we sample the parametric domain Ω with a regular
Figure 7 shows an example and provides a comparison between equilateral triangle mesh M 2D that defines the base structure of our
our spatially graded auxetic linkage and the homogeneous pattern linkage. The user selects the resolution and orientation of this mesh
proposed in [Konaković et al. 2016]. Note that the nonuniform link- to match her design intent. Lifting M 2D onto S by the inverse map
age structure no longer fully opens or closes in the plane like the reg- f −1 yields M 3D .
ular auxetic linkage could; once any region (hexagonal opening) in
3D Linkage Optimization. We now obtain an initial guess for the
the pattern fully opens or fully closes, further expansion/contraction
fully-opened linkage structure by constructing the medial triangle
requires spatially varying the stretch factors, inducing curvature
for each triangle in M 3D (i.e., inscribing a triangle by connecting
that forces the structure into 3D.
edge midpoints; see Figure 8). While this initialization is already
close to the desired target configuration, the discrete nature of the
5 MATERIAL OPTIMIZATION lifting function introduces inaccuracies that necessitate further opti-
In this section, we describe our computational workflow and the op- mization. In particular, we need to ensure that the linkage triangles
timization algorithm for computing the deployable auxetic linkage remain equilateral and are maximally expanded everywhere while
for a given design surface. staying close to the target surface. Fortunately, these objectives
can be formulated easily in the context of the projective approach
Preprocessing. Our first step is to analyze the input surface to of [Bouaziz et al. 2012]. Specifically, to obtain the linkage’s curved
ensure that it satisfies the positive mean curvature requirement. target configuration L 3D we minimize an energy function E L3D de-
As discussed in Section 3.4, we correct infeasible surfaces by ap- fined as the sum of three different objective terms over the vertex
plying mean curvature flow adapted to operate only on regions of positions x,
non-positive mean curvature(see also Figure 9). We use implicit
integration for the flow as proposed by [Desbrun et al. 1999]: E L3D (x) = ω 1 E expand (x) + ω2 E equi (x) + ω 3 E design (x), (2)
t t t +1 t t
(M + hL )x =M x ,
with weights ωi . Each term can be formulated as a sum of constraint
where M is the mass matrix, h is the step size, L is the positive proximity functions of the form ϕ(xc ) = ∥xc − P(xc )∥22 , where xc is
semidefinite cotan Laplace matrix, x is a matrix of vertex positions the vertex set involved in the specific constraint, and P denotes the
(one row per input surface vertex), and the superscripts indicate the projection operator to the constraint set, as detailed below.

ACM Trans. Graph., Vol. 37, No. 4, Article 106. Publication date: August 2018.
 Rapid Deployment of Curved Surfaces via Programmable Auxetics • 106:9

We observe that in the fully expanded where (i, j) denotes the vertex indices of an edge and E is the set of
state, the hexagonal openings formed by edges of the linkage. The operator
the linkage must attain maximum area. By
∥xi − xj ∥
Cramer’s theorem [Niven 1981, p. 236], this P E (ui , uj ) = (ui − uj )
maximum is achieved when all vertices of ∥ui − uj ∥
the opening lie on a circle.
projects to the closest edge with target length ∥xi − xj ∥ of the
We thus introduce the expansion term
Õ corresponding edge in the 3D linkage L 3D . We also add the non-
E expand = ∥xh − PC (xh )∥22 , penetration constraint proposed in [Konaković et al. 2016] to avoid
h ∈H collisions in the 2D state. The final optimized linkage L 2D then
defines the flat auxetic surface material that deploys to the desired
where h is an index set of vertices in a particular hexagonal opening,
target state.
and H is the collection of all such index sets in the linkage. PC (xh )
defines the projection to the circle closest to the vertices of xh Algorithm Parameters. Our implementation of the projective con-
computed as described in [Bouaziz et al. 2012]. straint solver is based on the open-source library provided by [Deuss
Contrary to the uniform pattern used in [Konaković et al. 2016], et al. 2015]. We set the weights in (2) to ω 1 = ω 2 = 100 and ω 3 = 1
our linkage triangles need to vary in scale to introduce spatially and apply between 100 to 600 iterations, depending on the mesh
varying maximal expansion. In order to let triangles scale freely but resolution. Total computation time for 3D and 2D optimization for
keep their equilateral shape, we introduce the energy a linkage with 8k triangles is 1.8 minutes on a standard desktop
Õ computer with 4.2 GHz computed on a single core.
E equi = ∥xt − PT (xt )∥22 ,
t ∈T Cone Singularities. When the conformal scale factors exceed the
where t is the index set of the vertices of a triangle, T is the set of all maximal expansion limits of the auxetic linkage, we need to insert
linkage triangles, and PT is the projection to the closest equilateral cone singularities in the conformal map to reduce scale distortion.
triangle, computed using shape matching as described in [Umeyama Singularities can also be mandated by the input surface’s topology
1991]. (to satisfy the Gauss-Bonnet theorem). These singularities corre-
Finally, to keep the linkage close to the design surface, we apply spond to boundary vertices of M 2D where the incident boundary
positional constraints of the form curves (seams) close up when lifted to M 3D by the conformal map.
Õ Because conformal maps preserve an-
E design = ∥xv − PS (xv )∥22 , gles, for the surface to close up and form
v ∈V a regular equilateral triangle mesh when
where v is a vertex index, V is the set of all linkage vertices, and PS lifted to M 3D , the sum of triangle an-
defines the projection to the closest point on S. gles around the singular vertex in M 2D —
The minimization of E L3D then follows the typical local/global referred to as the cone angle—must be an
iteration strategy (see also [Sorkine and Alexa 2007]): the local integer multiple of π3 . In the inset figure,
step computes all the constraint projections involved in the objec- we show an example with cone angle 5π 3
tive terms for the fixed current vertex positions; the global step and see how the equilateral triangle mesh (and an inscribed linkage)
subsequently solves for the optimal vertex positions keeping the will properly stitch together when lifted to M 3D . Figure 10 shows
constraint projections fixed. Details on the precise definitions of examples with singularities of cone angle 4π 5π
3 and 3 .
the projection operators and the corresponding numerical solver
implementations can be found in [Bouaziz et al. 2012] and [Deuss In-plane Opening. In case the computed scale factors do not fully
et al. 2015]. cover the maximal admissible range, the resulting 2D linkage can
still be expanded in the plane until one hexagonal opening is fully
2D Linkage Optimization. The 3D optimization provides us with opened—or contracted until one opening is fully closed—as shown
the curved target configuration L 3D of the linkage in its fully opened in the inset.
state. Now we need to find the contracted linkage in the plane that We leverage this property for the
defines the material rest state to be fabricated. We formulate this fabrication result in Figure 1 to re-
problem as a second projective optimization. We first apply the duce the material stresses at the tri-
necessary topological cuts to convert M 2D into a regular triangular angle joints during inflation by pre-
linkage L 2D with uniform triangle sizes (Figure 8). Note that this flat opening the linkage as much as pos-
linkage has a one-to-one vertex correspondence with the deployed sible; this minimizes the rotation
linkage L 3D . Next, we optimize the 2D vertex coordinates u of L 2D necessary to achieve the fully ex-
so that the triangles assume the edge lengths of L 3D . This is again panded configuration. In the opti-
easily implemented using a projective edge length constraint of the mization, we add an additional angle
form constraint [Deng et al. 2015] with a low weight that either tries to
Õ
E edge = ∥(ui − uj ) − P E (ui , uj )∥22 , expand or contract the linkage in the flat configuration, depending
(i, j)∈E on the user’s preference.

ACM Trans. Graph., Vol. 37, No. 4, Article 106. Publication date: August 2018.
106:10 • Konaković-Luković et. al.

Fig. 9. From left to right: input design surface, modified surface with positive mean curvature everywhere, optimized linkage in deployed state, 2D rest state
of flat-fabricatable material. In the bottom row, a singularity of cone angle 5π
3 is introduced to bring the conformal scale factors to lie within the admissible
range.

Filling in the Surface. If the user desires a deployed surface with- of positive mean curvature. In the smooth setting, our analysis in
out holes, the hexagonal openings in the fully expanded linkage can Section 3 shows that this defines a deployable target surface under
be filled in by layering four sheets offset from each other: inflation or gravity. To verify that this observation also holds in
the discrete case, and that the computed linkage does indeed define
a steady state under inflation or gravitational loading, we apply a
physics-based simulation. We use the same projective approach as
we did for linkage design, using only edge length constraints to keep
triangles rigid and positional constraints to fix the boundary. We
augment this optimization with dynamics as proposed in [Bouaziz
However, simply creating copies of the optimized linkage L 3D and et al. 2014] by applying forces on the linkage vertices. For inflation,
shifting them does not work: this would effectively translate the the force vectors are oriented along the surface normal, for gravity
deployed surface itself and also would lead to triangles imperfectly along the fixed negative vertical axis.
fitting the hexagonal holes due to the varying scale factors. In- Our experiments confirm that the linkages properly deploy, reach-
stead, these sheets must be designed by offsetting copies of M 2D in ing an equilibrium configuration very close to L 3D . For each of our
the parametric domain and lifting/optimizing them in 3D. Figure 4 examples, we compute the maximal distance between vertices in
shows an example of a surface filled in with this method. L 3D and their corresponding pairs in the equilibrium linkage. We
then compute a maximum relative vertex deviation for each model
Verification by Simulation. Recall that the above optimization by dividing this distance by the length of the bounding box diagonal.
maximizes surface expansion of the linkage on a target surface

ACM Trans. Graph., Vol. 37, No. 4, Article 106. Publication date: August 2018.
 Rapid Deployment of Curved Surfaces via Programmable Auxetics • 106:11

no information about the target shape, which is solely encoded in

the linkage pattern computed by our algorithm. Note that while the
inflated surface has positive mean curvature everywhere as required
by our analysis, both positive and negative Gaussian curvature are
present in the target shape. Figure 10 shows how cone singulari-
ties are introduced when inflating closed surfaces. The structure
of Figure 11 illustrates how gravitational forces can be used to de-
ploy the target shape from the flat rest configuration. The resulting
structure is in tension everywhere and can thus be used for physical
form-finding of self-supporting structures in analogy to the famous
approach of Antoni Gaudi [Fernandez 2006].

Application Case Studies. Figure 2 illustrates an application in

personalized medicine, where a deployable freeform coronary stent
can be customized to a specific patient. The stent is fabricated as
a flat structure, then rolled into a thin cylinder. When inflated,
the stent adopts the desired freeform shape to best advance blood
flow in the critical artery. Figure 3 highlights an application in
deployable architecture to construct a relocatable, semi-permanent
structure. Compared to the simple geometries of existing inflatable
structures, our approach supports a broader class of freeform shapes,
which allows adapting the structure to the design-specific interior
space requirements. Figure 4 illustrates how multiple layers of our
programmable auxetic material combine to create the approximately
closed surfaces of a freeform chair (deployed by gravity).

7 LIMITATIONS AND FUTURE WORK

Our current deployment strategies using inflation or gravity can
only actuate a subset of the surfaces realizable with a graded aux-
Fig. 10. Cone singularities are required when modeling closed surfaces etic linkage: those with positive mean curvature. Adding additional
such as the sphere. The simulated models at the top depict examples with constraints—for example, in the form of strings connecting cer-
cone angles of 4π /3 and 5π /3, respectively. At the bottom, lower-resolution tain vertices and thus preventing expansion towards positive mean
fabricated prototypes with cone angles of 4π /3 are shown, one fabricated curvature—can enlarge the space of deployable shapes. It is an in-
by laser cutting with triangles connected by rings, one 3D printed with
teresting question for future work to find a minimal set of such
ball joint connections. The surfaces have been closed manually along the
boundary elements prior to inflation.
constraints for a given target surface.
For closed surfaces and surfaces requiring singularities, we must
introduce cuts to flatten the material to the plane (see Figure 10 and
The model with the worst relative deviation has a maximal vertex bottom row of Figure 9). While this retains the benefits of planar
distance of 0.0513 and a bounding box diagonal of 28.5, giving a fabrication, the deployment becomes more complex, as the material
relative deviation of 0.0018. In all cases, the differences are nearly has to be re-connected along the cut seam prior to actuation.
imperceptible. Our results confirm that we obtain a close approximation of the
target shape even for relatively coarse resolutions of the linkage.
6 RESULTS However, although we have observed several connections to confor-
mal mapping theory that inform our optimization algorithms, we
We verify our material design and optimization approach with a
currently do not have a discrete theory for the geometry of graded
number of numerical and physical experiments, illustrating a range
auxetic linkages. Developing such a theory in the context of discrete
of potential application fields with different materials and usage
differential geometry is an exciting avenue for future work.
domains. Figure 9 shows several examples of our deployable auxetic
While our fabricated prototypes provide a proof-of-concept for
surfaces computed with the optimization algorithm described in
the physical realizability of our designs, we do not address impor-
Section 5.
tant fabrication-related issues at different scales. In particular, it is
Fabricated Prototypes. Figure 1 shows how one can deploy a crucial for robust deployment to optimize the joints connecting the
doubly-curved freeform surface from a single flat sheet of mate- linkage triangles. It also would be interesting to test techniques for
rial. The expansive forces for deployment are created by a generic permanently rigidifying the deployed structure. We hope that our
rubber balloon that is inflated against the support plane. As the bal- work can stimulate new research in material science, mechanical
loon is pumped with air, it presses against and deforms the linkage engineering, and architectural construction to study these questions
until the target shape is reached at maximal stretch. The balloon has in more detail.

ACM Trans. Graph., Vol. 37, No. 4, Article 106. Publication date: August 2018.
106:12 • Konaković-Luković et. al.

Fig. 11. Deployment via gravity. The auxetic linkage shown on the bottom right has been optimized to match the input design surface on the left. The structure
has been assembled in the flat state from individually laser-cut triangles that are connected by metallic rings to enable the rotational motion of the linkage
triangles. When lifted onto the rectangular support, the surface automatically deploys into its target shape. Note that boundary vertices are fixed along the
long edges of the support rectangle, and connected with strings on the short edges.

8 CONCLUSION REFERENCES
Numerous physical objects, such as ship or airplane hulls, build- Bernd Bickel, Moritz Bächer, Miguel A. Otaduy, Hyunho Richard Lee, Hanspeter Pfister,
Markus Gross, and Wojciech Matusik. 2010. Design and Fabrication of Materials
ing facades, clothing, and many consumer products are fabricated with Desired Deformation Behavior. ACM Trans. Graph. 29, 4, Article 63 (July 2010),
by shaping thin, initially planar materials. The shaping process 10 pages.
Sofien Bouaziz, Mario Deuss, Yuliy Schwartzburg, Thibaut Weise, and Mark Pauly. 2012.
typically involves bending, stretching, or otherwise deforming the Shape-Up: Shaping Discrete Geometry with Projections. Comput. Graph. Forum 31,
material using a mold or scaffold to guide the deformation towards 5 (2012), 1657–1667.
the desired 3D shape. Deployable structures provide an alternative Sofien Bouaziz, Sebastian Martin, Tiantian Liu, Ladislav Kavan, and Mark Pauly. 2014.
Projective Dynamics: Fusing Constraint Projections for Fast Simulation. ACM Trans.
where the shaping process and resulting target geometry are implic- Graph. 33, 4, Article 154 (July 2014), 11 pages.
itly encoded in the structure itself. We have shown that spatially Isaac Chao, Ulrich Pinkall, Patrick Sanan, and Peter Schröder. 2010. A Simple Geometric
graded auxetics are well suited to implement deployable surface Model for Elastic Deformations. ACM Trans. Graph. 29, 4, Article 38 (July 2010),
6 pages.
structures. Instead of rationalizing a 3D design surface for a given Xiang Chen, Changxi Zheng, Weiwei Xu, and Kun Zhou. 2014. An Asymptotic Numer-
homogeneous material, we spatially optimize the material itself. By ical Method for Inverse Elastic Shape Design. ACM Trans. Graph. 33, 4, Article 95
(July 2014), 11 pages.
carefully controlling the expansion behavior of the material, we di- Fionnuala Connolly, Conor J. Walsh, and Katia Bertoldi. 2017. Automatic design of
rectly program the target surface geometry into the flat 2D rest state. fiber-reinforced soft actuators for trajectory matching. Proceedings of the National
Inflation or gravitational loading then automatically deploys the Academy of Sciences 114, 1 (2017), 51–56.
Bailin Deng, Sofien Bouaziz, Mario Deuss, Alexandre Kaspar, Yuliy Schwartzburg,
rest state towards the target, which is assumed when the material and Mark Pauly. 2015. Interactive Design Exploration for Constrained Meshes.
cannot expand any further. As a consequence, we can leverage the Computer-Aided Design 61 (2015), 13–23.
efficiency of 2D digital fabrication technologies without requiring Mathieu Desbrun, Mark Meyer, Peter Schröder, and Alan H. Barr. 1999. Implicit Fairing
of Irregular Meshes Using Diffusion and Curvature Flow. In Proceedings of the 26th
any additional 3D guide surface. Our deployment strategy is robust Annual Conference on Computer Graphics and Interactive Techniques (SIGGRAPH ’99).
and reversible, which supports efficient storage and transport and ACM Press/Addison-Wesley Publishing Co., New York, NY, USA, 317–324.
Mario Deuss, Anders Holden Deleuran, Sofien Bouaziz, Bailin Deng, Daniel Piker, and
enables new applications for semi-permanent structures. Mark Pauly. 2015. ShapeOp—A Robust and Extensible Geometric Modelling Paradigm.
The combination of limited-expansion auxetic material with a Springer International Publishing, 505–515.
deployment via inflation or gravity imposes limits on the space of Günay Doğan and Ricardo H. Nochetto. 2012. ESAIM: Mathematical Modelling and
Numerical Analysis 46, 1 (2012), 59–79.
realizable shapes. Our analysis clearly delimits this space and di- Levi H. Dudte, Etienne Vouga, Tomohiro Tachi, and L. Mahadevan. 2016. Programming
rectly informs our computational solution, providing designers with curvature using origami tessellations. Nature Materials 15, 5 (2016), 583–588.
an effective tool to realize new deployable structures not possible J. Eells and J. C. Wood. 1976. Restrictions on Harmonic Maps of Surfaces. Topology 15
(1976), 263–266.
before. Santiago Fernandez. 2006. Structural Design in the Work of Gaudi. 49 (12 2006),
324–339.
Jan Friedrich, Sven Pfeiffer, and Christoph Gengnagel. 2018. Locally Varied Auxetic
ACKNOWLEDGMENTS Structures for Doubly-Curved Shapes. Springer Singapore, Singapore, 323–336.
C.J. Gantes. 2001. Deployable Structures: Analysis and Design. WIT Press.
We are grateful to Pavle Konaković for his help with design, fab- Mohammad Ghomi. 2017. Open Problems in Geometry of Curves and Surfaces. http:
//people.math.gatech.edu/~ghomi/Papers/op.pdf. (Oct. 2017). Accessed: 2017-12-12.
rication, and renderings. We thank the CMU Geometry Collective Ruslan Guseinov, Eder Miguel, and Bernd Bickel. 2017. CurveUps: Shaping Objects
members and Gaspard Zoss for insightful discussions. We also thank from Flat Plates with Tension-actuated Curvature. ACM Trans. Graph. 36, 4, Article
Liane Makatura, Peng Song, and the anonymous reviewers for 64 (July 2017), 12 pages.
Caigui Jiang, Chengcheng Tang, Amir Vaxman, Peter Wonka, and Helmut Pottmann.
their valuable feedback. This work was supported by the NCCR 2015. Polyhedral Patterns. ACM Trans. Graph. 34, 6, Article 172 (Oct. 2015), 12 pages.
Digital Fabrication, funded by the Swiss National Science Founda- Sun-Jeong Kim, Soo-Kyun Kim, and Chang-Hun Kim. 2002. Discrete differential error
metric for surface simplification. In 10th Pacific Conference on Computer Graphics
tion, NCCR Digital Fabrication Agreement #51NF40-141853, NSF and Applications, 2002. Proceedings. 276–283.
Award 1717320, and gifts from Autodesk Research and Adobe Re- Mina Konaković, Keenan Crane, Bailin Deng, Sofien Bouaziz, Daniel Piker, and Mark
search. Pauly. 2016. Beyond Developable: Computational Design and Fabrication with

ACM Trans. Graph., Vol. 37, No. 4, Article 106. Publication date: August 2018.
 Rapid Deployment of Curved Surfaces via Programmable Auxetics • 106:13

Auxetic Materials. ACM Trans. Graph. 35, 4, Article 89 (July 2016), 11 pages.
W. Y. Lam. 2017. Minimal surfaces from infinitesimal deformations of circle packings.
ArXiv e-prints (Dec. 2017).
Ying Liu, Brandi Shaw, Michael D. Dickey, and Jan Genzer. 2017. Sequential self-folding
of polymer sheets. Science Advances 3, 3 (2017).
Li-Ke Ma, Yizhong Zhang, Yang Liu, Kun Zhou, and Xin Tong. 2017. Computational
Design and Fabrication of Soft Pneumatic Objects with Desired Deformations. ACM
Trans. Graph. 36, 6, Article 239 (Nov. 2017), 12 pages.
I. Niven. 1981. Maxima and Minima Without Calculus. Number v. 6 in Dolciani
Mathematical Expositions. Mathematical Association of America.
Jifei Ou, Mélina Skouras, Nikolaos Vlavianos, Felix Heibeck, Chin-Yi Cheng, Jannik
Peters, and Hiroshi Ishii. 2016. aeroMorph - Heat-sealing Inflatable Shape-change
Materials for Interaction Design. In Proceedings of the 29th Annual Symposium
on User Interface Software and Technology (UIST ’16). ACM, New York, NY, USA,
121–132.
Johannes T. B. Overvelde, Twan A. de Jong, Yanina Shevchenko, Sergio A. Becerra,
George M. Whitesides, James C. Weaver, Chuck Hoberman, and Katia Bertoldi. 2016.
A three-dimensional actuated origami-inspired transformable metamaterial with
multiple degrees of freedom. 7 (11 03 2016).
Julian Panetta, Qingnan Zhou, Luigi Malomo, Nico Pietroni, Paolo Cignoni, and Denis
Zorin. 2015. Elastic Textures for Additive Fabrication. ACM Trans. Graph. 34, 4,
Article 135 (July 2015), 12 pages.
Jesús Pérez, Miguel A. Otaduy, and Bernhard Thomaszewski. 2017. Computational
Design and Automated Fabrication of Kirchhoff-plateau Surfaces. ACM Trans. Graph.
36, 4, Article 62 (July 2017), 12 pages.
Jesús Pérez, Bernhard Thomaszewski, Stelian Coros, Bernd Bickel, José A. Canabal,
Robert Sumner, and Miguel A. Otaduy. 2015. Design and Fabrication of Flexible Rod
Meshes. ACM Trans. Graph. 34, 4, Article 138 (July 2015), 12 pages.
Ahmad Rafsanjani and Damiano Pasini. 2016. Bistable auxetic mechanical metamaterials
inspired by ancient geometric motifs. Extreme Mechanics Letters 9 (2016), 291 – 296.
Dan Raviv, Wei Zhao, Carrie McKnelly, Athina Papadopoulou, Achuta Kadambi, Boxin
Shi, Shai Hirsch, Daniel Dikovsky, Michael Zyracki, Carlos Olguin, Ramesh Raskar,
and Skylar Tibbits. 2014. Active Printed Materials for Complex Self-Evolving Defor-
mations. 4 (18 12 2014).
Rohan Sawhney and Keenan Crane. 2017. Boundary First Flattening. ACM Trans. Graph.
37, 1, Article 5 (Dec. 2017), 14 pages.
Krishna Kumar Saxena, Raj Das, and Emilio P. Calius. 2016. Three Decades of Auxetics
Research - Materials with Negative Poisson’s Ratio: A Review. Advanced Engineering
Materials 18, 11 (2016), 1847–1870.
Alexander Schiftner, Mathias Höbinger, Johannes Wallner, and Helmut Pottmann. 2009.
Packing Circles and Spheres on Surfaces. ACM Trans. Graph. 28, 5, Article 139 (Dec.
2009), 8 pages.
Christian Schumacher, Bernd Bickel, Jan Rys, Steve Marschner, Chiara Daraio, and
Markus Gross. 2015. Microstructures to Control Elasticity in 3D Printing. ACM
Trans. Graph. 34, 4, Article 136 (July 2015), 13 pages.
Mélina Skouras, Bernhard Thomaszewski, Bernd Bickel, and Markus Gross. 2012. Com-
putational Design of Rubber Balloons. Comput. Graph. Forum 31, 2pt4 (May 2012),
835–844.
Mélina Skouras, Bernhard Thomaszewski, Peter Kaufmann, Akash Garg, Bernd Bickel,
Eitan Grinspun, and Markus Gross. 2014. Designing Inflatable Structures. ACM
Trans. Graph. 33, 4, Article 63 (July 2014), 10 pages.
Olga Sorkine and Marc Alexa. 2007. As-rigid-as-possible Surface Modeling. In Pro-
ceedings of the Fifth Eurographics Symposium on Geometry Processing (SGP ’07).
Eurographics Association, 109–116.
Hideshi Tomita, Takashi Higaki, Toshiki Kobayashi, Takanari Fujii, and Kazuto Fuji-
moto. 2016. Stenting for curved lesions using a novel curved balloon: Preliminary
experimental study. 66 (8 2016), 120–124. Issue 2.
Shinji Umeyama. 1991. Least-Squares Estimation of Transformation Parameters Be-
tween Two Point Patterns. IEEE Trans. Pattern Anal. Mach. Intell. 13, 4 (1991),
376–380.
Amir Vaxman, Christian Müller, and Ofir Weber. 2017. Regular Meshes from Polygonal
Patterns. ACM Trans. Graph. 36, 4, Article 113 (July 2017), 15 pages.
Changxi Zheng, Timothy Sun, and Xiang Chen. 2016. Deployable 3D Linkages with
Collision Avoidance. In Proceedings of the ACM SIGGRAPH/Eurographics Sympo-
sium on Computer Animation (SCA ’16). Eurographics Association, Aire-la-Ville,
Switzerland, Switzerland, 179–188.
Bo Zhu, Mélina Skouras, Desai Chen, and Wojciech Matusik. 2017. Two-Scale Topology
Optimization with Microstructures. ACM Trans. Graph. 36, 5, Article 164 (July 2017),
16 pages.

ACM Trans. Graph., Vol. 37, No. 4, Article 106. Publication date: August 2018.