Conference PaperPDF Available

Introduction to Playable Voxel-Shape Grammars


Abstract and Figures

A shape grammar is a collection of visually defined geometric rules that could be used to automate the generation of formal representations of designs for buildings, cities, products and more. We offer an extension of the shape grammar formalism based entirely on voxel space instead of vectors, which we used for the generation of schematic architectural designs. We describe a method using playability to increase human agency and designer control over the outcome of the generative phase of voxel-shape grammars. The method is presented with an implementation in the environment of Minecraft and employs three guidance mechanisms. To conclude we list a few considerations from our experience in the design of a playable, voxel-shape grammar and point to future work.
Content may be subject to copyright.
Introducon to Playable
Voxel-Shape Grammars
1 Some voxel shapes in the language
dened by the voxel-shape
grammar shown in Figure 4, as
experienced by the users of the
grammar in Minecra. In the fore-
ground, an acvaon state for rule
Anton Savov
DDU — Digital Design Unit
TU Darmstadt
Oliver Tessmann
DDU — Digital Design Unit
TU Darmstadt
A shape grammar is a collecon of visually dened geometric rules that could be used to auto-
mate the generaon of formal representaons of designs for buildings, cies, products and more.
We oer an extension of the shape grammar formalism based enrely on voxel space instead
of vectors, which we used for the generaon of schemac architectural designs. We describe a
method using playability to increase human agency and designer control over the outcome of the
generave phase of voxel-shape grammars. The method is presented with an implementaon in
the environment of Minecra and employs three guidance mechanisms. To conclude we list a few
consideraons from our experience in the design of a playable, voxel-shape grammar and point to
future work.
The ability of shape grammar generave formalism to capture
design intenons and expert knowledge is what interests us in
the pursuit of engaging non-experts in the architectural design
process. We base our work on this large eld of research and
present a novel extension to shape grammars, namely playable,
voxel-shape grammars that could have an enabling eect upon
both experts and non-experts by leng them interacvely and
immersively explore design opons and learn by making. With
our shape grammar method we target use scenarios in the early
schemac design phase, when all stakeholders explore the rela-
onship between the elements of a project in an informal and
loose manner, oen in the form of massing models.
From very early on in the work we aimed to:
1. Oer user-friendly visual modelling of both the shapes in
the shape grammar, as well as the grammar rules, in order to
aord the expert users easy access to the tooling framework—
this was possible by redening the typically vector-based
shape grammars for a voxel space and using Minecra as the
implementaon environment.
2. Open the generave process to human agency in an inter-
acve, manageable and surmountable way—by making the
grammar playable, we could add direct parcipaon and
guiding mechanisms to the otherwise purely algorithmic
generave process.
We developed and tested an extendable framework imple-
menng playable, voxel-shape grammars in the Minecra
environment. This paper presents:
a denion of playable, voxel-shape grammars (Secon 3.1);
test cases at urban and building scales (Secon 4);
three types of guiding mechanisms that experts could use to
control the play-driven generave process (Secon 5.3).
Shape grammars have been successfully used to encode aspects
of expert architectural knowledge, and thus help their users,
mostly designers, to explore a large set of design soluons or
quickly generate two-dimensional representaons of designs
with desired formal qualies (Sny and Gips 1972; Sny and
Mitchell 1978; Chase 2002). The applicaon of shape grammars
consists of mostly two acvies:
1. designing the grammar, which consists of shapes and rules;
2. iterang mulple mes through the rules to generate designs.
The second acvity maintains the usefulness of shape grammars
as a design tool, subject to their implementaon as automated
systems run by a computer (Ruiz-Monel et al. 2013). However,
shape grammars have been mostly vector-based (Sny 1980;
Woodbury 2016), and the shapes generated with grammars are
only notaonal graphs or abstract 2D composions. (Marn
2006; Sny and Mitchell 1978; Duarte 2005). This poses two
main challenges in the aempt to use them as design tools:
1. computer implementaons of shape grammars are dicult to
2. computer-automated generaon of shapes with shape gram-
mars is dicult to explore and control.
Computer Implementations of Shape Grammars are
Difficult to Define
An important aspect of a successful shape grammar implemen-
taon is that the users are able to model the shapes and rules
Usually, shape grammars have been dened in a closed, expert
coding environment with an interface dened solely for the
purposes of the shape grammar. Therefore, a potenal new user
of the shape grammar cannot rely on computer skills they already
have, and must learn new, advanced ones. The use of generic,
accessible and easy-to-use 3D-modelling environments to imple-
ment shape grammars has not been well explored. The following
18-year-old quote from Terry Knight states the problem rather
well, and unfortunately not much has happened since then to
address it:
Currently though, there are few computer implementaons that
are praccable for students or praconers. Most do not have
interfaces that make them easy for nonprogrammers to use.
More eorts have gone to computaonal problems than to inter-
face ones. Implementaons of simple, restricted grammars that
are visual and require only graphic, nonsymbolic, nonnumerical
input are needed. (Knight 1999)
We address this by implemenng our method for playable,
voxel-shape grammars in a popular and accessible 3D environ-
ment with more than 40 million worldwide users of all ages and
backgrounds: the game Minecra. The size of a user’s Minecra
avatar in relaon to the voxel shapes they model—and subse-
quently iterate through in the game world—creates an immersive
percepon for a one-to-one architectural scale.
Computer-Automated Generation of Shapes with Shape
Grammars is Difficult to Explore and Control
Most exisng research on shape grammars uses a computer-au-
tomaon approach to the generave phase, leaving no control,
or very lile, mostly global control, to the designer during the
process of shape generaon. This has been done “as pencil-
and-paper execuon might be tedious and therefore useless for
the sake of discovering many new soluons.” (Ruiz-Monel et al.
2013). However, an automated generaon process is most oen
opaque and non-iterable—the user cannot inuence the result
while the process runs and is simply presented with an outcome.
Our method uses a combinaon of playability and guiding mech-
anisms to make the iteraon process more navigable.
George Sny (1980) denes a vector shape as follows: "A shape
is a limited arrangement of straight lines dened in a cartesian
coordinate system with real axes and an associated euclidean
metric.” In a similar manner let us dene voxel shape as a limited
arrangement of voxels in a discrete voxel grid and an associated
euclidian metric (Figure 3).
A voxel-shape will dene all further terms we nd in Sny's
“Introducon to Shape and Shape Grammars” (1980), but for the
discrete voxel space instead of connuous cartesian vector space.
For example, a scale operaon on a voxel shape needs to have
an integer number for the scale factor—for example 2x—so that
it turns one voxel into a cube of eight voxels: 2 x 2 x 2. Rotaons
would go only in 90 degree increments and so on. A less strict
denion, allowing real number scale factors and rotaon, is
also possible, but it would need a post-processing voxel detailing
roune, such as marching cubes, and thus introduce unnecessary
complexity for the current purposes of the method (Marais and
Crumley 2012).
The original shape grammars formalism uses vectors that are
simple representaons of architectural elements such as walls.
Because of its 3D nature, the proposed voxel-shape grammar
could be used to represent both elements (walls, slabs) as well as
massing volumes of a building’s or city’s elements.
An important aspect of shape grammars is the labelled shape,
consisng of shape and symbols, used as a matching mask for
the grammar rules. The marker symbols in vector space shapes
are notaonal. In our voxel shapes we use voxels with special
aributes for markers (Figure 3).
Shape grammars perform computaons with shapes in two steps:
a recognion of a parcular shape and its possible replacement.
Rules specify the parcular shapes to be replaced and the
Savov, Tessmann
2 The original graphic denion of
shape grammars as it appeared in
the 1980 paper “Introducon to
Shape and Shape Grammars.
3 A vector shape as dened by Sny
and Gips 1980 and a voxel shape as
dened in this research.
4 Graphic denion of a sample
voxel-shape grammar and some
of the shapes in the language it
5 Sample playable voxel-shape
grammar (le: rules, right: inial
shape)—besides the voxels’ state,
the rules take into account also the
player's posion.
6 Generaon of a voxel shape using
the playable voxel-shape grammar
shown in Figure 5.
7 A guiding mechanism provides a
desirable advantage to the players
for their next moves and thus could
inuence their decisions.
8 The urban scale voxel-shape
grammar from IBA_GAME has rules
that place pre-designed buildings.
9 A schemac design for a residenal
building created using a voxel-
shape grammar with 6 rules.
manner in which they are replaced. (Willis 2012)
A shape grammar is dened by rules and an inial shape, and is in
essence a language of shapes. Shape grammar rules consist of a
mask shape on the le and a replacing shape on the right (Figure
For the shape rules’ implementaon in voxel space we dene a
voxel shape as a mask, containing at least one marker voxel and a
replacing voxel shape (Figure 4). A voxel-space 3D version of the
2D vector-shape grammar from Figure 2 is seen on Figure 4.
The use of shape grammars in voxel space has only been previ-
ously explored on two occasions. Marais and Crumley (2012)
presented an extension to shape grammars called voxel-space
shape grammars, in which the shapes and rules are dened in
vector form and only the generated shapes—aer the iteraon is
complete—are voxelized in a post-processing step. A more appro-
priate naming for their method would be voxelized, vector-shape
grammars. Our method is therefore sll novel, as the enre
grammar is dened and iterated in voxel space.
Friedman and Stamos (2013) introduce the noon of a voxel
grammar in their work on procedural tree generaon. However,
their grammar denion is not visual but numerical, and as such,
lacks many of the advantages of the original shape grammar
formalism which “… allows for algorithms to be dened directly
in terms of […] shapes” (Sny 1980). Our implementaon here
suggests that the enre voxel-shape grammar (shapes and rules)
is dened in a visual, user-friendly way.
As such, voxel-shape grammars would be easy to model but
would have all the problems of shape grammars dened in
secon 2.
To address the control and ease-of-use issue described in 2.2, we
introduce human agency in the form of game mechanics. It adds
one addional step in the generave iteraon of the grammar
rules—a player needs to place some of the markers in order to
complete a mask shape found in the rules (Figure 5). Besides
the voxel state, the rules in the playable version of voxel-shape
grammars also take into account the player posion.
Diering from shape grammars, where the generaon is auto-
mated and each step of the iteraon produces the condions
for the next, playable voxel-shape grammars require that every
iteraon needs to be iniated by a player, thus making conscious
choice (i.e., a design decision) an important part of the explora-
on process (Figures 1, 5, and 6).
Savov, Tessmann
On the other hand, player decisions could be inuenced by a
guidance mechanism. A guiding mechanism is dened by the
author of the grammar and is able to encode a design intenon
or expert knowledge. A well-dened guiding mechanism would
be integrated in the game and would in essence be a means to
provide a desirable advantage to the players for their next moves,
and thus could inuence their decisions (Figure 7).
We are tesng the concept of playable voxel-shape grammars
in the project 20.000 BLOCKS. Our implementaon consists of
a Minecra version with the ComputerCra mod and a set of
custom scripts which drive rule detecon, player/goal tracking
and results export. See Savov, Buckton, and Tessmann (2016)
and for further details on the project.
We invesgated the use of playable voxel-shape grammars at
two scales. At the urban scale the shapes in the grammar repre-
sent pre-designed whole buildings (or mass models of buildings),
and the rules represent the possibilies for these buildings
to exist next to each other in a city. We tested this approach
between August 2016 and March 2017 in the IBA_GAME—a
20.000 BLOCKS version that we created for IBA Heidelberg,
where players can create small neighbourhoods. Hundreds of
these neighbourhoods form a new quarter for the German city
of Heidelberg. The game featured a voxel-shape grammar of 40
buildings of six types: houses; housing extensions; businesses;
science and tech spaces; gathering spaces; and public plazas
(Figure 8). It had 140 players who played it 820 mes. The shape
grammar rules went through about 10 major revisions in the
course of development.
10 Variaons based on shape rotaons for the HOUSE RULE (rule 2 from Figure 8) in the urban scale grammar.
At the building scale the shapes in the grammar could repre-
sent physical elements of a building, such as walls and slabs or
the massing of the volume a room or an area would take in the
building, while the rules represent the logic for those elements
to aggregate next to each other to form a building. To test this
approach we ran a SmartGeometry workshop in April 2016
focusing on residenal, mul-apartment buildings. Three teams
of architects and engineers used our implementaon of playable
voxel-shape grammars and dened components of the buildings
within the grammar (Figure 9). Each variaon had up to 16 shape
rules and was played by aendees of the conference.
Based on our two-year experience with this ongoing project, we
oer the reader the following thoughts on voxel shapes, voxel-
shape rules and guidance mechanisms.
The abstracted modelling environment with a voxel size of 1 x 1
x 1 m could be liming. However, since our aim is to generate
schemac architectural designs, this level of abstractness works
in our favour.
An important requirement in order to achieve an immersive
experience in the generave phase—i.e., in the play phase—is
that the resulng structures are walkable for players, so they can
reach the newly added markers, otherwise the iteraon process
is blocked. That means that in both urban or building scales, all
possible sequences of the grammar’s shape rules need to create
a walkable combinaon when placed.
Color Coding
The points of possible further grammar growth need to be very
clearly visible to the player while in play. Therefore, the textures
of the marker voxels and the building blocks need to be dierent.
Furthermore, with larger sets of rules, such as in IBA_GAME,
where we used six types of rules with 2–4 variaons each, a
color coding of the types of rules helps the player learn the
grammar faster.
Rotation Options
As we are operang in a strict orthogonal voxel grid, we had four
possible rotaons for each shape rule in the grammar. We found
out that modelling them to be the same in behaviour, but only
similar in looks, decreases the rigidity of designs and introduces
visual complexity (Figure 10).
Thus far, we have not used scaling or rotang of the shapes, i.e.,
the masking shapes that are modelled in the grammar rules are
absolute in size and orientaon. Due to the discrete quality of
a voxel space, scaling and rotaon in free increments could be
detrimental to the readability of a shape’s design.
Features of the Rule System
Markers Placed by the Grammar
We used two types of marker voxels:
1. the acvator block on which a player steps to acvate the
detecon of a masking shape they had built around it (Figure
2. the extension block, which is a marker that replaces the ac-
vator block upon successful detecon.
Some of the grammar rules require the extension block in their
masking shapes. The extension block ensures that structures
grow out of themselves coherently instead of populang the
design space with too many scaered, independent struc-
tures. That means some shapes in the grammar act as seeds,
for example the shape represenng a house in IBA_GAME,
and others as extensions, for example the shapes represenng
commercial buildings in IBA_GAME (Figure 8).
Markers Placed by the Player
Player-placed marker voxels, i.e., resources, are in essence the
material with which the player “draws” or models the masking
shape. For example in IBA_GAME, we used two resources that
encode a set of meta design features into the voxel shapes:
1. an urban resource -> shapes are more solid, with vercally
proporoned windows and enclosed;
2. a green resource -> shapes are lighter, with horizontally
proporoned windows and more open (Figure 11).
When using both types of resources to build the same masking
shape, the player would expect the same type of resulng shape,
with the same behavior during the iteraon process, but with
dierent design qualies.
Markers in a Grid or Freely Spaced
When designing the playable voxel-shape grammar, the expert
needs to make sure that all rules are spaally compable with
each other and create a walkable design. Walkability can be
obstructed by unforeseen shape overlaps.
One strategy we enforced in the rules is growth along, or within,
the constraints of a three-dimensional grid. The markers were
always placed in a shape according to a grid of 9 x 9 x 8 blocks
(Figure 10). We also developed an in-house design style guide
Savov, Tessmann
to regulate, across our team, how and when secons of a shape
rule could overlap with another (Figure 12). This saved the work
of having to try all combinaons separately.
Another strategy, which avoids the rigidity of using a grid, is for
the shapes to come with the markers in a posion that allowed
for the making of stepped structures and richer designs in spaal
terms. To avoid too many unwanted overlaps, new structures
only grow from exisng ones (seeds), to allow us to detect rule
adjacency (Figure 9).
Features of the Guidance Mechanisms
We implemented playability in the form of an economy system:
if a player owns the resources and the territory needed to create
a masking shape, then they could acvate a grammar rule with it
and so gain the rewards given by the rule. We used playability as
a generave force and at the same me as a control mechanism
over the outcome. Ruiz-Monel et al. (2013) make the case for
control mechanism as means to guarantee feasible designs.
For the designer to be able to encode a design intenon into the
playable voxel-shape grammars, we tried three types of guidance
mechanisms: a quantave game-goal system; liming choice
with the shape design; and performance-based feedback.
Quantitative Guidance Mechanism
We used a so goal system that qualitavely guided the growth
as dierent players chose to go for dierent goals. For example
in IBA_GAME we dened four server-wide goals: collect the
most green points, build the most buildings, build the tallest
structure, build as many dierent buildings as possible (Figure
11 Two resources that describe a
set of metadesign features of the
voxel-shapes in the grammar. Le:
the green resource and the house
associated with it. Right: an urban
resource with its house variaon.
12 An extract from the design style
guide for the urban scale grammar
of IBA_GAME, which was devel-
oped and shared within our team.
13 Liming player choices with design.
Le: With a rule that allows users
to stack the same house over itself,
the most ecient way for a player
to increase their height score is to
stack them up. Right: If the voxel
shape, by design, prevents direct
stacking, the next most ecient
way of gaining height points is to
criss-cross houses.
14). Each voxel-shape rule in the grammar rewards the player
dierently on all four metrics.
Limiting Choice with the Shape’s Design
If players pursue a so goal such as height, then they will aim
to build as high as possible with as lile eort and expense as
possible. To prevent the stacking of one and the same house on
top of itself, we designed the roof of the house so that players
cannot model the masking shape for a new house on top of it
(Figure 13).
Performative Guidance Mechanism
While sll in the early works in terms of seamless integraon, we
tested a guiding system based on structural analysis. It exports
the model to Grasshopper every me a rule is acvated, runs an
analysis roune on it, and then “reports” back the results to the
players by color-coding certain blocks of the structure (Savov,
Buckton, Tessmann 2016). We used a similar approach in the
project Sensive Assembly, documented in Savov, Tessmann, and
Nielsen (2016), where this type of feedback loop is described in
more detail.
We presented an extension to the shape grammar formalism
for voxel spaces, which we used for the generaon of schemac
architectural designs. We described a method using playability to
increase human agency and designer control over the outcome
of the generave phase of voxel-shape grammars. At the end we
shared a few thoughts on the design of a playable voxel-shape
grammar in Minecra.
Potential Contribution to the Field
With our method architects can encode architectural logic and
principles in the grammars and easily modify them. The gram-
mars are volumetric and thus open to immediate interpretaon
and analysis for their architectural potenal without further
transformaon or translaon steps.
Furthermore, the method allows for the integraon of design and
fabricaon. An addional reason to use voxel-shape grammars
instead of vector-shape grammars is to come one degree closer
to embedding material specicaon rules in the grammar (Rossi
and Tessmann 2017; Sny and Gips 1971).
The method relies on players using the Minecra map to model
grammars and generate structures. Currently we have had 800
plays, 150 play structures, and 150 downloads. There are the
following bolenecks to increasing parcipaon:
1. The need to download and use a customized Minecra
version, which reduces the number of potenal players. We
are looking for ways to implement the grammar logic enrely
on the server side so that players can parcipate with a
Vanilla (unmodied) Minecra client.
2. Closed server access: currently the data collecon requires us
to run the server ourselves. We are exploring opons to let
users run their own Minecra server with the map and save
results in the cloud.
Future Work
The research could benet from the following future
A generave visualizer, which could simulate a game being
played so that a designer can capture some of the most
common conict/bugs in advance, like mismatching shapes
and broken walkability, without having to play the game over
and over again.
Implemenng automated rules that would not need the user’s
input at every generave step. There needs to be a beer
balance between the eects of the player’s intenons and
the generave rules over the nal designs. To address this, in
some cases, when the right condions occur, the grammar
rules might allow for several iteraon steps to happen auto-
macally one aer the other, in the form of a bonus eect or
what Sanchez (2015) denes as synergies.
The collected data (grammar denions plus the generated
shapes) can be used to train a machine learning algorithm to
generate either grammar sets or schemac designs.
The test game can be downloaded at:
The project 20.000 BLOCKS is led by Anton Savov and developed at the
DDU (Digital Design Unit, Prof. Dr. Oliver Tessmann) at the architectural
faculty of Technische Universität Darmstadt with the following team:
Steen Bisswanger, Ben Buckton, Theo Gruner, Jörg Hartmann, Thomas
Valenn Klink, Sebasan Koerer, Marios Messios, Lorena Müller, Max
Rudolph, Alexander Stefas, Alban Voss. The IBA_GAME version of 20.000
BLOCKS was made possibly through the partnership with IBA Heidelberg
and sponsored by Eternit GmbH.
Chase, Sco C. 2002. “A Model for User Interacon in Grammar-Based
Design Systems.Automaon in Construcon 11 (2): 161–72.
Crumley, Zacharia, Patrick Marais, and James Gain. 2012. “Voxel-Space
Shape Grammars.” In Proceedings of the 20th Internaonal Conference on
Computer Graphics, Visualizaon and Computer Vision, edited by Vaclav
Skala, Part I, 113–22. Plzen, Czech Republic: WSCG.
Duarte, José P. 2005. “A Discursive Grammar for Customizing Mass
Housing: The Case of Siza’s Houses at Malagueira.Automaon in
Construcon 14 (2): 265–75.
Friedman, Sam, and Ioannis Stamos. 2013. “Automac Procedural
Modeling of Tree Structures in Point Clouds Using Wavelets.” In
Proceedings of the 2013 Internaonal Conference on 3D Vision, 215–22.
Seale, WA: 3DV.
Knight, Terry. 1999. “Applicaons in Architectural Design, Educaon and
Pracce.Report for the NSF/MIT Workshop on Shape Computaon. Vol. 2.
Cambridge, MA.
Marn, Jess. 2006. “Procedural House Generaon: A Method for
Dynamically Generang Floor Plans.” In Symposium on Interacve 3D
Graphics and Games. Redwood City, CA: I3D.
Rossi, Andrea, and Oliver Tessmann. 2017. “Designing with Digital
Materials.” In Protocols, Flows, and Glitches: Proceedings of the 22nd
Annual Conference of the Associaon for Computer-Aided Architectural
Design Research in Asia, edited by P. Janssen, P. Loh, A. Raonic, and M.A.
Schnabel, 37–42. Suzhou, China: CAADRIA..
Ruiz-Monel, Manuela, Javier Boned, Juan Gavilanes, Eduardo Jiménez,
Lawrence Mandow, and José-Luis Pérez-de-la-Cruz. 2013. “Design with
Shape Grammars and Reinforcement Learning.” Advanced Engineering
Informacs 27 (2): 230–45.
Sanchez, Jose. 2015. “Block’hood - Developing an Architectural
Simulaon Video Game.” In Real Time: Proceedings of the 33rd Internaonal
Conference on Educaon and Research in Computer Aided Architectural
Design in Europe, vol. 1, edited by Bob Martens, Gabriel Wurzer, Thomas
Grasl, Wolfgang El. Lorenz, and Richard Scharanek, 89–97. Vienna:
Savov, Anton, Ben Buckton, and Oliver Tessmann. 2016. “20,000 Blocks.”
Savov, Tessmann
In ACADIA // 2016: Posthuman Froners: Data, Designers, and Cognive
Machines, Proceedings of the 36th Annual Conference of the Associaon
for Computer Aided Design in Architecture, edited by Kathy Velikov, Sean
Ahlquist, Maas del Campo, and Georey Thün, 24–33. Ann Arbor:
Savov, Anton, Oliver Tessmann, and Sg Anton Nielsen. 2016. “Sensive
Assembly: Gamifying the Design and Assembly of Facade Wall
Prototypes.Internaonal Journal of Architectural Compung 14 (1): 30–48.
Sny, George, and W. J. Mitchell. 1978. “The Palladian Grammar.
Environment and Planning B: Planning and Design 5 (1): 5–18.
Sny, George. 1980. “Introducon to Shape and Shape Grammars.
Environment and Planning B 7 (3): 343–51.
Sny, George, and James Gips. 1972. “Shape Grammars and the
Generave Specicaon of Painng and Sculpture.” In Proceedings of
Informaon Processing 71, Vol. 2, 1460–65. Amsterdam: IFIP.
Willis, Bryce R. 2012. “Digital Morphogenesis: A Computaonal Housing
Typology.” MA Thesis, University of Nebraska, Lincoln.
Woodbury, Robert. 2016. “An Introducon to Shape Schema Grammars.
Environment and Planning B: Planning and Design 43 (1): 152–83.
Figure 2: George Sny, 1980
All other drawings and images by the authors.
Anton Savov works as a research associate at the Digital Design Unit
(DDU) — TU Darmstadt and is the founder of architectural studio
AWARE. He iniated the plaorm “20.000 Blocks" as a means for
collaborave architectural design and successfully partnered with IBA
Heidelberg to create the IBA_GAME based on it. Previously, Anton has
worked at Bollinger+Grohmann Ingenieure and taught at the Städelschule
Architecture Class. His work has been exhibited at the Venice Biennale,
MAK Center in Los Angeles, NODE Frankfurt and others. If you want to
get in touch, email to
Oliver Tessmann is an architect and Professor for Digital Design at TU
Darmstadt. His research and teaching is located in the eld of computa-
onal design and digital fabrica on. A er receiving his doctorate from the
University of Kassel in 2008, he served as the director of the performa-
veBuildingGroup at Bollinger+Grohmann. From 2010 to 2012 he was Visi
ng Professor at the Städelschule Architecture Class in Frankfurt and, from
2012 to 2015, held the posi on of Assistant Professor at KTH Stockholm.
14 Quantave guidance mechanism in the form of four goals: greenest, tallest, densest, most diverse neighbourhood.
... Most contributions in this cluster were related to the voxel model applications for visualizing large datasets describing buildings [53] and large territories [54]. Experiments with preliminary design exercises both on-screen [55][56][57] and in virtual reality [58,59] exist. Voxel-based generative-design interfaces have also been proposed [60,61]. ...
... Strehlke [55] Savov and Tessmann [57] De Klerk et al. [58] Fischer [60] Erioli and Zomparelli [61] Asmar [70] Thariyan [84] Leder [127] Xiao [128] Michalatos and Payne [132] Morello et al. [69] Mitasova et al. [101] Design Development Cubukcuoglu et al. [71] Gorte et al. [72] Breslav and Khan [73] Wang et al. [77] Baron et al. [135] Mekki et al. [136] Ambrozkiewicz and Kriegesmann [137] Aage et al. [139] Materials and Manufacturing/ Construction ...
Full-text available
Although voxel models have been applied to address diverse problems in computer-aided design processes, their role in multi-domain data integration in digital architecture and planning has not been extensively studied. The primary objective of this study is to map the current state of the art and to identify open questions concerning data structuring, integration, and modeling and design of multi-scale objects and systems in architecture. Focus is placed on types of voxel models that are linked with computer-aided design models. This study utilizes a semi-systematic literature review methodology that combines scoping and narrative methodology to examine different types and uses of voxel models. This is done across a range of disciplines, including architecture, spatial planning, computer vision, geomatics, geosciences, manufacturing, and mechanical and civil engineering. Voxel-model applications can be found in studies addressing generative design, geomatics, material science and computational morphogenesis. A targeted convergence of these approaches can lead to integrative, holistic, data-driven design approaches. We present (1) a summary and systematization of the research results reported in the literature in a novel manner, (2) the identification of research gaps concerning voxel-based data structures for multi-domain and trans-scalar data integration in architectural design and urban planning, and (3) any further research questions.
... This makes grammars too complex to define and navigate, especially when defined in 3D space. Two notable exceptions are the work by Rossi and Tessmann on the discrete modeling library WASP, which is based on graph grammars and offers voxelized scalar fields to navigate the generation process and the playable voxel-shape grammars presented by Savov and Tessmann who discretized the shape grammars method into a voxel grid and made both the rule definition and the iteration process a game-like experience (Rossi & Tessmann 2018, Savov & Tessmann 2017. The Wavefunction Collapse Algorithm (WFC), a type of constraint-solving algorithm, is widely used in the gaming industry for its ease of use and graphical nature (Karth & Smith 2017). ...
... As described in section 3, a purely automated generation process is most often opaque and non-iterable-the user cannot influence the result while the process runs and is simply presented with an outcome (Savov & Tessmann 2017). On the other hand, for a non-expert it is not easy to advance in a design task if not assisted by a system that offers a large influence on the design, some kind of automated intelligence where expertise is encoded. ...
Full-text available
Developing participatory computational design systems for a wider professional community of architects is a key enabler to explore the benefits of massively-collaborative creative design search and discovery. In this paper we formulate three principles to increase the creative control of non-expert users when developing participatory computational design systems. We present an overview of four existing strategies for encoding design content into digital modeling environments and introduce a fifth one, assisted sculpting, that uses iso-surfacing and constraint-solving. Finally, we describe our technical implementation of an assisted sculpting digital environment with potential applications in creating massing models and schematic designs.
... Crumley (2012) introduced voxelization in shape grammar for reducing the computational complexity of meshbased implementation. Building on this initial research, Savov and Tessmann (2017) developed playable voxel grammar in Minecraft. The work described here extends the study of voxel grammar to explore their application in formal analysis, where a precise definition of spatial and programmatic relations is desired. ...
... Most of these games are focused on the morphological aspect of the built environment. In the architectural scale, the exploration of complex spatial configuration can be structured as a voxelbased grammar in [156], or a shape grammar with structural verification routines in [157]. On the urban scale, rule-based systems have been used extensively to describe the morphological relations of the city [158]. ...
Full-text available
This chapter provides a methodological overview of generative design in architecture, especially highlighting the commonalities between three separate lineages of generative approaches in architectural design, namely the mathematical optimization methods for topology optimization and shape optimization, generative grammars (shape grammars and graph grammars), and [agent-based] design games. A comprehensive definition of generative design is provided as an umbrella term referring to the mathematical, grammatical, or gamified methodologies for systematic synthesis, i.e. derivation, itemization, or exploration of configurations. Among other points, it is shown that generative design methods are not necessarily meant to automate design but rather provide structured mechanisms to facilitate participatory design or creative mass customization. Effectively, the chapter provides the theoretical minimum for understanding generative design as a paradigm in computational design; demystifies the term generative design as a technological hype; shows a precis of the history of the generative approaches in architectural design; provides a minimalist methodological framework summarising lessons from the three lineages of generative design; and deepens the technological discourse on generative design methods by reflecting on the topological constructs and techniques required for devising generative systems or design machines, including those equipped with Artificial Intelligence. Moreover, the notions of discrete design and design for discrete assembly are discussed as precursors to the core concept of design as decision-making in generative design, thus hinting to avenues of future research in manufacturing-informed combinatorial mass customization and discrete architecture in tandem with generative design methods.
... Within architectural research, approaches toward the implementation of software devoted to design with discrete units have been proposed in recent years, such as the implementation of discrete logics in game engines [27] or the use of voxel grammars [28]. Other approaches used RHINO and GRASSHOPPER as their base platform but have often not been released to the general public, or have been released for use, but not as open-source packages for further development. ...
Full-text available
This article summarises a series of interconnected researches exploring the potential of applying topological interlocking methodologies to the field of architectural design and fabrication. Specifically, it describes two concurrent approaches to design with interlocking units, the first relying on parametric design logics and mass customized fabrication processes, and the second implementing discrete combinatorial processes for both design and fabrication using modular units. We first outline the historical background of combinatorial thinking in architectural computing, and describe the emergence of computational design and digital fabrication. We further present the recent evolution of a combinatorial design paradigm, which challenges the acquired parametric design methodologies in computational architecture research. We then present our research in the field of modular design and topological interlocking, presenting the transition of the research from parametric to combinational logics. We describe design and fabrication methodologies for both approaches, and evaluate the potentials and limitations of both. We further present recent work in the development of software for combinatorial design within CAAD software, and its first applications to design of topological interlocking systems. We conclude by outlining the future research directions and possibilities of integration between parametric and combinatorial processes in design, fabrication and assembly of interlocking systems.
... Within architectural research, construction systems inspired by digital materials logics aim at defining novel strategies for building adaptation and transformation. Approaches towards the implementation of software devoted to design with discrete units have been proposed in the recent years, including the use of game engines [19] and voxel grammars [20]. All these approaches rely on custom developed solutions, and as such, lack potential for scalability and usability for users not trained in programming. ...
Full-text available
Developing from researches on heterogeneous modelling and voxel-based design, as well as on digital materials and discrete assembly, the paper describes current developments of Wasp, a software tool for hierarchical discrete modelling of objects as aggregations of modular parts, with focus on the generation of architectural objects. By conceptualizing the discrete nature of the final objects already within the software, it becomes possible to model heterogenous artefacts composed of basic parts, which are reversibly joined into a complete aggregation. Within such framework, users are provided with different aggregation procedures to select rule sequences producing the desired outcomes, as well as basic utilities for part collisions avoidance, visualization and editing of the produced geometries. To increase control over designed objects, the framework has been extended to include hierarchical modelling with multi-resolution parts, more advanced generation and editing functions for voxel fields, multi-channel scalar fields and global constraints. Overall, the aim of the framework is to allow modelling of objects as aggregations of discrete units with reversible connections, hence allowing the production of architectural entities which could be assembled, disassembled, and reconfigured during their lifecycle.
Based on generative adversarial networks (GANs), a type of deep learning model, this paper takes the functional layout of the emergency departments (EDs) of general hospitals as the research object, combines the hierarchical design concepts and proposes an intelligent functional layout generation method for EDs. It aims to explore the application of intelligent algorithms in architectural design and build an intelligent design method to solve the generation problem of ED layouts. The specific process of this method is as follows: First, 120 sets of ED drawings with excellent layouts are collected and labelled. Second, three of the most representative GAN frameworks including deep convolutional generative adversarial network (DCGAN), image-to-image translation with conditional adversarial network (pix2pix) and cycle-consistent adversarial network (CycleGAN) are chosen to establish training models for the ED layout. Finally, the rationality of the generated results is analysed from an architectural perspective, while the loss functions and trend for the generator and discriminator are compared from the algorithmic perspective. The analysis of the three GANs’ results shows that these models can autonomously generate new ED function layouts, of which the DCGAN results are the most flexible but the image quality is not ideal. The pix2pix outputs have the highest image quality, but the dataset has strict constraints. The CycleGAN has loose requirements on the dataset and yields ideal results with strong applicability. Some scholars have used pix2pix to explore the generation of apartment floor plans in recent years. However, this study established three different GAN models for hospitals for the first time, and compared their generation results to explore the applicability of different GANs.
Full-text available
The article describes a method for gamifying the design and assembly of computationally integrated structures built out of discrete identical blocks. As a case study, the interactive installation Sensitive Assembly was designed and built at the Digital Design Unit (Prof. Dr Oliver Tessmann) at the Technische Universität of Darmstadt and exhibited during the digital art festival NODE 2015 in Frankfurt in 2015. Sensitive Assembly invites people to play a Jenga-like game: starting from a solid wall, players are asked to remove and replace the installation’s building blocks to create windows to a nurturing light while challenging its stability. A computational system that senses the current state of the wall guides the physical interaction and predicts an approaching collapse or a new light beam breaking through. The installation extends the notion of real-time feedback from the digital into the physical and uses machine-learning techniques to predict future structural behaviour.
Conference Paper
Full-text available
We present a method for discovering the structure of trees in 3D point clouds by linking wavelets with shape grammars. Given a range scan of a tree we find a grammar that can reproduce that tree, and others like it, with sub-voxel accuracy. The grammar inferred is stochastic, allowing us to generate many permutations of related trees. The method of multi-resolution analysis, employed by the discrete wavelet transform, gives great insight into tree structure. Trees are self-similar and exhibit similar branching patterns at different resolutions. The wavelets make these patterns explicit by decomposing the tree into different levels of detail. The multi-resolution structure of the wavelet transform also allows us to infer an L-System grammar. The productions in the grammar are derived from the progressive levels of refinement in the wavelet transform. Each production maps a vector in the low resolution image to a set of vectors in the higher resolution image. Our method utilizes the Fast Wavelet Transform opening the door to real-time inference of procedural models. The grammar inferred is concise and generative, allowing for compression and graphics applications of our algorithm. We demonstrate novel applications of the grammar for shape completion, scan enhancement and geometry propagation.
Full-text available
Shortly after shape grammars were invented by Stiny and Gips, a two part project for shape grammars was outlined by Stiny. In a 1976 paper,1 Stiny described "two exercises in formal composition". These simple exercises became the foundation for the many applications of shape grammars that followed, and suggested the potential of such applications in education and practice. The first exercise showed how shape grammars could be used in original composition, that is, the creation of new design languages or styles from scratch. The second exercise showed how shape grammars could be used to analyze known or existing design languages. Both exercises illustrated the unique characteristics of the shape grammar formalism that helped motivate a quarter century (almost!) of shape grammar work. General but simple, formal yet intuitive: qualities that continue to make shape grammar disciples and confound skeptics. The history of shape grammar applications in architecture and the arts for the two complementary purposes of synthesis and analysis, as well as for a third, joint purpose is sketched in the first section of this report. These three categories of applications do not have rigid boundaries. They are used in this report mostly as a framework for discussion. An overview of the roles of shape grammar applications in education and practice is given in the second section. New and ongoing issues concerning shape grammars in education and practice are discussed in the last section.
Architects have integrated computers into firms to streamline the documentation process and which has allowed for the integration of rapid prototyping and digitally driven technologies and tools. Although this has increased the efficiency of the traditional approach to architecture, an alternative methodology has the potential to adapt the computer’s role in architecture, making it a more integrated part of the design process. Within a traditional process, software allows a designer to build the documentation of his designs around the relationships between elements. Instead, new methodologies can be used to imbed the nature of an architectural design within a system of internal parametric representations. (Yessios) This allows for the creation of computationally designed systems where an interactive framework could be used to aid in the design process. This paper discusses parametric design method being used to generate housing based on site constraints, typological features, and pragmatic housing functions and details. The current home-building market is led by developers who consider custom residential architecture to be a list of interior finishes from which a home buyer can choose. As a result, four or five floor plans populate a neighborhood. Architects currently account for a negligible portion of the residential architecture industry, being limited primarily to the design of rare, expensive custom homes. Although these homes often push the typology of a residential architecture, they are not an economical solution for home design. In the paper “Towards a Fully Associative Architecture,” Bernard Cache showcases the Philibert De L’Orme Pavilion and his fully associative design and manufacturing process which allowed him to produce everything from the initial form of the pavilion to the 100percent CNC custom kit of parts. Cache’s projects elaborate on the traditional design methodology and production methods. The parametric methods he employs are used to define greater complexity within his designs. Like most custom homes designed by architects, these methods are not widely affordable, and so the power of parametric methods cannot be captured for average people. By implementing new methodologies these underutilized parametric systems can be leveraged to generate custom home solutions to both fully engage computers within an architectural design process and raise the quality of current housing practices.
Shape schema grammars generalize parametric shape grammars so that both rules and the objects to which they apply are expressed with shape schemata. This paper defines shape schema grammars. It starts with a notation for schemata in general and shape schemata in particular. Schema equality is shown to have at least three possible definitions, of which schema consistency is the most useful. A limited notion of shape schema maximality potentially reduces the size and redundancy of a given schema. Shape schema subpart is a multifunction returning all of the possible ways that one shape schema can be embedded in another. Shape schema difference and addition complete the basic mathematical operations over shape schemata required to define shape schema rules, grammars and languages.
Shape grammars are a powerful and appealing formalism for automatic shape generation in computer-based design systems. This paper presents a proposal complementing the generative power of shape grammars with reinforcement learning techniques. We use simple (naive) shape grammars capable of generating a large variety of different designs. In order to generate those designs that comply with given design requirements, the grammar is subject to a process of machine learning using reinforcement learning techniques. Based on this method, we have developed a system for architectural design, aimed at generating two-dimensional layout schemes of single-family housing units. Using relatively simple grammar rules, we learn to generate schemes that satisfy a set of requirements stated in a design guideline. Obtained results are presented and discussed.
Grammar-based production systems are considered potentially powerful design tools by their ability to generate sets of designs adhering to user-specified constraints. However, development of such tools has been slow, partly because of the lack of good interaction between user and system. This paper describes modes of user interaction and control possible with grammar-based design systems and presents issues to be examined in the development of models that represent the locus of interactions possible with such systems. The examination of existing grammar-based systems provides empirical evidence to support the validity of such models.
The ultimate goal of the described research is a process for mass customizing housing based on computer-aided design and production systems. The current goal is the development of an interactive system for generating solutions on the Web based on a modeling approach called discursive grammar. A discursive grammar consists of a programming grammar and a designing grammar. The programming grammar generates design briefs based on user data; the designing grammar provides the rules for generating designs in a particular style, and a set of heuristics guides the generation of designs towards a solution that matches the design brief. This paper describes the designing grammar using Siza's houses at Malagueira as a case study.