Automation in Construction 140 (2022) 104385

Contents lists available at ScienceDirect

Automation in Construction
journal homepage: www.elsevier.com/locate/autcon

Review

Automated floorplan generation in architectural design: A review of

methods and applications
Ramon Elias Weber *, Caitlin Mueller, Christoph Reinhart
Building Technology Program, Department of Architecture, Massachusetts Institute of Technology, 77 Massachusetts Avenue, Cambridge, MA 02139, USA

A R T I C L E I N F O A B S T R A C T

Keywords: Accommodating predicted population growth and urbanization within the UN Climate Goals poses a significant
Floorplans challenge for disciplines that engage with the built environment. High performing buildings of the future should
Automation offer spatial quality for their users while utilizing resources as efficiently as possible for both construction and
Generative design
operation. In this review, we survey the value proposition of automatic floorplan layout generation methods and
Buildings
their opportunities for design guidance, feedback, and optimization in the creation of new buildings, in addition
Sustainability
Layout optimization to applications for inventory characterization to survey existing housing stock and guide building policy and
Machine learning code. We divide existing methods into three categories: bottom-up methods, top-down methods, and referential
Geometry methods. We explore advantages and challenges for each approach and propose a hybrid method for future
building layout automation that utilizes a new set of metrics to create sustainable buildings of the future.

1. Introduction width or depth requirements. In terms of future reuse opportunity, the

design team would ideally also like to know how amenable a given
The International Energy Agency (IEA) estimates that the global built floorplan is to adaptive reuse or which walls could be load bearing while
floor area will increase by some 235 billion m2 until 2050, to accom­ supporting good daylighting etc.
modate a growing population and rising standards of living [1]. The This paper reviews previous automated floorplan layout generation
environmental, economic, and ethical implications of this prediction are methods and assesses their actual and potential application for archi­
momentous. At a time when humanity has around 580 Gt CO2e left to tectural design, urban planning, and real-estate development. A floor­
burn to keep global warming below 1.5 ◦ C, the “plan” to double the size plan layout creation method creates an architectural layout from a series
of the building sector in a single generation seems risky. While one may of geometric constraints and/or programmatic requirements. It should
expect significant efficiency gains in terms of area-weighted resource already be noted here that thus far only experimental artistic architec­
use and construction costs vis-à-vis today’s building practices, the tural design practices [2,3] and the real-estate sector have been eager to
crucial question is whether humanity really needs that indoor space? embrace novel generative and artificial intelligence-based layout auto­
The construction sector – and especially the high-performance design mation tools, whereas the architectural profession at large has expressed
community – have long embraced computer-based performance analysis some reservation against algorithms whose perceived ultimate goal
methods for embodied and operational energy use associated with might be to replace the profession. The purpose of this paper is to clarify
construction materials and building use. However, space efficiency the capabilities and limitations of existing methods and envision how
evaluations are far less common. Usually, there is a design brief provided they could contribute towards the design of more elegant, effective, and
to the architect that stipulates a certain amount of program including a flexible spaces on a scale ranging from individual floors and buildings to
set percentage for circulation and space conditioning equipment. The whole cities.
sum of these space uses adds up to an overall building volume that can A number of reviews on computational layout automation have been
then be explored via massing studies. The spatial relationship between a conducted that included industrial facility layouts [4], focused on spe­
massing volume or a floorplan and the distribution of program is, of cific computational methods, such as evolutionary algorithms [5,6] or
course, quite complex, ranging from desired adjacencies to minimal agent based methods [7]. Furthermore methods have been surveyed

* Corresponding author at: Building Technology Program, Department of Architecture, Massachusetts Institute of Technology, 77 Massachusetts Avenue, Cam­
bridge, MA 02139, USA
E-mail address: [email protected] (R.E. Weber).

https://doi.org/10.1016/j.autcon.2022.104385
Received 21 March 2022; Received in revised form 17 May 2022; Accepted 23 May 2022
Available online 2 June 2022
0926-5805/© 2022 Elsevier B.V. All rights reserved.
R.E. Weber et al. Automation in Construction 140 (2022) 104385

with a focus on methods optimizing for energy usage [8,9]. In this During the late 1970s, George Stiny started working on the analysis
manuscript, we initially review which professions have thus far used and reproduction of building layouts via so-called parametric shape
automated space layout methods and for what purpose. We then intro­ grammars [12]. By understand the design and spatial qualities of
duce a new approach that productively combines these methods with a existing buildings such as Palladian villas [13] or Frank Lloyd Wright’s
specific focus on early-stage architectural massing studies and existing prairie houses [14], the underlying patterns could be deployed to
building stock characterization. recreate buildings of the same type. Similarly, Christopher Alexander
represented spatial relationships in traditional architectural floorplans
2. Value proposition via relational graphs and tree structures [15]. These abstractions were
essential first steps necessary for automating the generation of floor­
The first automated floorplan generation methods were introduced plans and continue to play a foundational role in contemporary
close to half a century ago. The underlying motivation has changed over computational approaches.
time and still varies significantly between different projects and tools Grammar based design methods have been successfully implemented
today. This section provides an account of significant historic precedents into computational workflows to procedurally compute building vol­
followed by contemporary use cases, in which the authors see exciting, umes with detailed facades. Notably Esri’s City Engine [16] which in­
new applications of the underlying technologies. tegrates the CGA++ shape grammar language [17] that enables the
generation of differentiated building envelopes for visualization pur­
poses of urban design proposals.
2.1. Past and current approaches As of today, automated building-level layout tools have not made
much headway into mainstream architectural practice where, their use
Automatic space layout creation methods were introduced as artistic is mainly reserved for speculative design exercises or specific niche
speculations on the future role of computers and artificial intelligence in applications such as office furnishing and electric lighting layouts in
architecture in the 1970s. Yona Friedman proposed the Flatwriter [10] to interior design [18] or complex programming exercised for hospitals,
generate apartment layouts that would accommodate usage preferences airports or large scale residential and commercial developments
of all neighbors in cooperative housing projects. Automating the layout [19–21]. For such applications, automatically generated design options
creation process was seen as an enabler for participatory design. Cedric can augment or replace conventional manual design processes by of­
Price proposed the “Generator”: A reconfigurable voxel based spatial fering not a single optimal solution but a family of directions for further
unit that could be reconfigured by visitors into different layouts [11]. design exploration [22].
Both pre-computational proposals envisioned floorplans designed by a In contrast to architectural design, the real estate sector has
modular kit of parts in a bottom-up process.

Table 1
Representative sample of contemporary automatic space layout creation methods in practice.
Typology Scale Output Client Use case Company Name Citation
Product

Residential, (L) Multiple Massing, architectural Real estate, architects Increase speed of design Software Archistar [23]
Commercial, Mixed buildings program
use
Res, Com, Mixed (L) Multiple Massing, floorplans Real estate Feasibility studies, real Software Testfit [24]
buildings estate
Residential (L) Multiple Massing, architectural Architects, Real estate N/A Software Matterlab [25]
buildings Program
Res, Com, Mixed (L) Multiple Massing, architectural Architects, Real estate Feasibility studies, Software Spacemaker [26]
buildings Program design exploration
Residential (L) Multiple Massing, architectural Construction company Modular construction Software KREO [27]
buildings Program
Res, Com, Mixed (L) Multiple Massing Architects, Real estate Sustainable design, Software Digital Blue [28]
buildings feasibility studies Foam
Res, Com, Mixed (L) Multiple Massing, architectural Architects Feasibility studies, Software Delve [29]
buildings Program design exploration
Res (S) Single floor Architectural layout Architects Feasibility Architecture finch 3d [30]
Educational (M) Single Architectural layout from Community Participatory design Government Seismic School [31]
building predefined building blocks App
Residential (L) Multiple Massing, architectural Community, architects, Participatory design, Government Prism App [32]
buildings Program real estate feasibility
Undefined (S-L) Building massing, lighting Architects, engineers Framework for Software Project [33]
layout, window placement automated design Refinery
Residential (L) Multiple Massing, architectural Architects, government Design exploration, Software Typenhaus+ [34]
buildings program community engagement
Undefined (L) Multiple Massing In-House design tool Design exploration Architecture Scout [35]
buildings
Undefined (L) Multiple Massing, architectural In-House design tool Design exploration Engineering Site Solve [36]
buildings program
Res, Com, Mixed (M) Single Massing, architectural Municipalities, real Design exploration, Software Omrt ostate [37]
building program estate, architects feasibility
Res, Com, Mixed (L) Multiple Massing Homeowners, real Land acquisition, real Software CityBldr [38]
buildings estate estate evaluation
Hotel (M) Single Massing, architectural Hospitality companies Feasibility, early-stage Software Parafin [39]
building program planning
Res, Com, Mixed (L) Multiple Massing, architectural In-house Design exploration, Architecture Gensler Blox [40]
buildings program feasibility
Residential (S) Single Furnished architectural Architects Increase speed of design Software PlanFinder [41]
Apartment layout

2
R.E. Weber et al. Automation in Construction 140 (2022) 104385

enthusiastically embraced and supported the creation of a plethora of dimensioned for a given floorplate and the influence of floor-plan ty­
floorplan and building automation software and practice. Several com­ pologies on the energy usage of a building [45] can give architects a
panies of varying size now focus on the creation of automatic layout more intuitive understanding in early design processes. Programmatic
tools to assist property developers and decision makers. They promise to changes based on different lighting and thermal requirements can have a
maximize the potential buildable area and perform automatic analysis of direct influence on a building’s energy budget: Interactive approaches
a site, creating semi-automatic feasibility studies that can inform in­ can give a more intuitive understanding of these requirements, leading
vestment opportunities for land acquisition and maximize rentable area. to solutions that can negotiate between requirements of different
Whether geared towards the real estate industry or conceived as in- stakeholders. Furthermore, materially integrated design processes can
house software tools in architecture and engineering firms, most cur­ visualize how different construction methods and material systems have
rent approaches tackle layout automation on the scale of a single different constraints during the design process. A direct comparison and
building massing. They include different apartment (or in the case of calculation of embodied carbon and achievable spans could help users
hospitality, hotel room) mixes and simplified core placements with find new more sustainable design solutions [46].
single or double loaded corridors.
Table 1 provides a snapshot of prominent automated space layout 2.4. Use case 2 - design guidance or optimization
creation method at the time of writing.
Generative layout tools can be used to augment different stages of
2.2. Future applications existing digital design workflows. Parametric design spaces can be
explored for optimization within predefined constraints [47], and
We note that there is currently no technical implementation of grammar- and aggregation-based automated methods have been used to
automatic space layout creation available that can fully replace a skilled create new types of modular structures [48]. As speculative and early-
human designer. More importantly, given the crucial role that archi­ stage design tools, automated approaches offer the opportunity to test
tecture can play to provide more socially equitable and environmentally ideas at scale and generate design options iteratively that would be
responsible spaces, we do not believe that the end goal of automated difficult to achieve with manual workflows (Fig. 2).
floor-plan generator methods should be to replace the human architect. With highly specific programmatic requirements in specialty typol­
Instead, we have identified three broad use cases that go beyond effi­ ogies, such as hospitals or airports, automated layout methods can help
ciency maximization and rather focus on improving the quality of the designers to optimize floorplans with adjacency, pathfinding, energetic
design process and resulting architecture. These use cases are design or daylight heuristics [8,19], or structural system efficiency [49]. Multi-
feedback, design guidance/optimization, and inventory objective optimization and objective functions that are highly specific to
characterization. the specified architectural problem can be used to negotiate between
different (competing or diverging) goals. A series of experimental hybrid
2.3. Use case 1 - design feedback semi-automated methods have been deployed in such design processes
where physics-based simulations can be steered by a user to inform
Building massing decisions can have a significant and hard to predict programmatic distribution of layouts [50].
impact on resulting interior space layouts, which in turn have cascading Referential automated methods can be deployed in later stages of
effects on building occupancy, structural efficiency, and even opera­ building design. Leveraging architectural catalogues and previously
tional energy use. Fast simulations and generative design tools can help generated designs, methods of automation can reuse and adapt estab­
designers develop their own intuition for such relationships. In struc­ lished design solutions for new problems. This has been successfully
tural engineering education for architects and engineers, real-time demonstrated on a material scale where algorithmic workflows can
simulations have become a useful tool to visualize problems and help identify closest fit solutions in existing material catalogues [51] as well
designers build a geometric intuition to create more efficient structures as for adaption of existing floorplan layouts into new building massing
[42,43]. Real-time visualization of the impact of design decisions can be [52].
useful to convey information to decision makers. When combined with
novel interfaces, non-expert stakeholders or the local community can be 2.5. Use case 3 – inventory characterization
engaged and learn about the design process more easily [44].
Plugging in to existing design workflows, automatic layout genera­ Automatic layout design tools not only offer opportunities for new
tion could help to visualize how changes in a building’s massing relates buildings, but could be used to characterize and redevelop existing
to constraints for circulation, program, or structural requirements, as urban environments. Making use of widely available geometric massing
illustrated in Fig. 1. Showcasing how building cores must be GIS datasets, existing building stock could be modeled on a building

Fig. 1. Pedagogical use of generative design tools to help build design intuition for circulation (left), program and energy (middle) and material-based con­
straints (right).

3
R.E. Weber et al. Automation in Construction 140 (2022) 104385

Fig. 2. Use of automatic methods as design tool for automatically populating building massing (left), optimizing existing building layouts (middle) and the creation
of novel design options (right).

level when combined with automatic floorplan layouts. This could lead review paper: Web of Science [54], Google Scholar [55], Journals
better and more detailed understanding of existing housing stock and its indexed in the Architecture and Civil Engineering disciplines from
embodied material quantities [46]. Scopus [56], as well as CumInCAD [57], a database of conferences and
A better and more visual understanding of future developments journals in the architectural computational design disciplines. In a sec­
enabled by current zoning could lead to more informed decisions for ond step we analyzed the references mentioned in the review and
housing policies and building laws and allow for non-experts from the methods articles as well as tracking novel work that cited the relevant
broader public to engage with planning processes. A clear understanding articles. A fully automated search and bibliometric analysis was not
of desired goals could lead to outcome or performance based zoning that possible as floorplan and layout automation keywords are used in
can take metrics such as urban comfort, mobility, and daylighting in to different disciplines for applications in electric circuit and factory layout
account [53]. Identifying possibilities of reuse or densification on a large planning and design. We identified 49 different methods with geometric
scale could empower lawmakers to better guide their cities development architectural outputs of which 14 are bottom-up, 27 are top-down, and 8
as depicted in Fig. 3. are referential. Methods with architectural intent that did not result in a
floorplan layout (e.g. only studied adjacency graphs in floorplans, or
3. Methods building massings) were excluded.

Following a description of possible use cases, this section reviews the 3.2. Bottom-up methods
computational methods underlying previously suggested approaches for
automatic space layout generators. With origins in various engineering Architectural design briefs often have highly prescriptive spatial
and computer science disciplines, many of the methods have been requirements. Because of heavily specified room sizes or adjacency
developed for different use cases and have been adapted for building constraints, layout designs often lend themselves well to be generated
design workflows. This opens the field to new ideas and approaches for via bottom-up design processes. Bottom-up generator methods are
spatial design, but can also lead to a mismatch, where methodologies therefore conceptually related to traditional design methods such as
have inherent shortcomings that are difficult to adapt to the re­ mind mapping of spatial relationships, bubble diagrams, and physical
quirements of the architecture and planning disciplines. We divide modelmaking strategies. When using modular construction logics that
existing approaches into three categories: bottom-up methods, top-down make use of prefabricated systems in concrete or timber [58] bottom-up
methods and referential methods. We outline the strengths and weak­ aggregation logics enable the exploration of different design variations
nesses of these categories vis-à-vis previously mentioned use cases. and part-whole relationships.
In a final structure, a series of predefined building blocks are
3.1. Indexing and search aggregated into a larger assembly. As computational methods for the
design of floorplans and building layouts, these aggregations can either
Four main databases were used to retrieve research articles for this be static (with predetermined architectural spaces) or adaptive

Fig. 3. Opportunities of automatic space layout tools to be utilized for the survey of existing housing stock (left), to explore opportunities of densification (middle)
and to identify implications of changes in zoning and building codes (right).

4
R.E. Weber et al. Automation in Construction 140 (2022) 104385

(changing during computation) and can be coupled with heuristics to building components or adjustments.
achieve a desired global outcome. Transformations during the aggre­
gation process occur on the individual parts themselves. 3.3. Top-down methods
During the bottom-up aggregation process, additional layers of in­
formation can be superimposed on the digital model to either change, Real-world architectural design is often highly constrained by pre­
swap out existing units or guide further aggregations. These heuristic defined building massing that stems from urban scale considerations,
methods can include analytic metrics such as spatial relationships building code, or regulations. This can result in highly prescriptive
(proximity requirements), environmental performance (daylight access, volumes that define the boundaries of a building that architects want to
energy usage), structural efficiency metrics, or geometric details (pro­ be fully occupied. When designing a building with such strong con­
portions). A schematic of a bottom-up automatic design process for a straints on the envelope, defined through contextual requirements, site
series of rooms is described in Fig. 4. Table 2 compares different ap­ boundaries, or the reorganization of an existing structure, top-down
proaches and implementations of bottom-up methods. design methods can be of interest. Methods for subdivision, fitting,
Bottom-up processes for exploratory and speculative design have shape packing, and iterative agent based methods have been applied
been embraced by the design community to create discrete building across architectural scales to automate design problems (Fig. 5), ranging
systems that reintegrate design thinking with computational methods of from the material scale with optimal placements and dimensioning of
design and means of production [72]. Applied to a building scale, they shell components [79] to the layout and partitioning of geographical
are particularly useful when designing for specific typologies that allow district scale [80].
for modular construction and design logics in their realization. Members Two promising technological inspirations and very active areas of
of a structure, the so-called discrete parts, can be aggregated to respond research originate from the VSLI layout design and the Facility Layout
to specific architectural and spatial constraints or construction re­ Problem (FLP). Working with hierarchical systems that have inter­
quirements, creating opportunities for robotic fabrication and recon­ connected rectangular modules, while integrating material constraints
figurable structures [73]. [81], the automation of VSLI circuits design has parallels to spatial
As a response to the large search spaces of bottom up design pro­ layouts. As an optimization problem from the engineering community
cesses, the Model Synthesis algorithm creates a set of custom constraints for arranging program in a given floor space, FLP is applied when ma­
that guide the aggregation of user defined modules into complex 3D chines in a factory hall for have to be laid out for a production line [82].
shapes [74]. Taking the adjacencies of parts of an existing 3D shape as an There has been significant interest in trying to transfer FLP methods to
input, the method can generate new variations of larger dimensions that the architectural domain to optimize the placement of room layouts.
satisfy the original constraints. The method, also referred to as Wave However, current methods for solving FLP problems make use of highly
Function Collapse (WFC), has since gained traction for creating 2D abstracted mathematical models that are difficult to be transferred to
textures (using two dimensional pixel adjacencies) and 3D models for real-world architectural environments and their implementation in
procedural level creation in computer games [75,76] as well as for available software tools on the market has been limited [83].
modular design in the architectural domain [77]. To further guide the Top-down methods take a massing or boundary as an input, as well
search towards solutions with controlled spatial qualities, machine as a series of entities as fillers or targets for insertion. The input design is
learning (ML) guided heuristics have been proposed [78]. subdivided based on geometric constraints to assign spaces. Compared
Methods of aggregation with large geometric freedom often create to the bottom-up method, the transformations are done on the global
large search spaces that need clever heuristics to guide the exploration boundary conditions directly, resulting in a solution that will always
and output of good results. Additionally, stochastic methods do not conform to the initial boundary condition. An overview of different top-
necessarily find a solution, based on the problem settings. Furthermore, down methods is given in Table 3.
the bottom-up methods make it difficult to embed and control layers of Heuristics can be used to assess the current state of subdivision and
hierarchy that are prevalent in architectural design such as different can inform next steps in the case of iterative optimization processes. This
levels of circulation or structural load transfer that require differentiated can be computed using mathematical programming, such as Mixed

Fig. 4. Schematic of bottom-up automatic space layout design methods. Starting with a definition of members and objectives, an initial aggregation and adjustment
transforms the individual parts themselves. An objective function evaluates the outcome and drives local and global parameters to adjust the outcome.

5
R.E. Weber et al. Automation in Construction 140 (2022) 104385

Table 2
Comparison of bottom-up automatic space layout creation methods in literature. Genetic Algorithms (GA), Mathematical programming (MP).
Typology Scale Objective Function Optimizer Inputs Output Speed Architectural Citation
Quality

N/A (S) Single Minimal wall length GA Number and areas of Floorplan, based on N/A Low [59]
floor rooms grid
Residential (S) Single Maximize cross GA Tree representation Floorplan, N/A Low [60]
floor ventilation, (perimeter to of program differentiated rooms
area ratio) and minimize connected
weighted sum of distances
(closeness of rooms)
Public (S) Single Alignment, adjacency, Physically Based Area, adjacency Modeling N/A Low [61]
floor orientation, proportion (of architectural design
single rooms) objectives in
physically based space
planning
Residential (S) Single Minimize gap space. Evolutionary Area, location Assigned program on N/A Low [62]
floor algorithm preference existing layout,
differentiated
boundaries
Residential (M) Connectivity, adjacency, 1.Baysian network Area, foootprint, Program layout ~ Seconds High [63]
Multiple envelope containment, for Program aspect ratio, to 7 min
floors convexity generation, 2. adjacency,
Metropolis algorithm adjacency type
Residential, (M) Shading of neighboring Quadratic Boundary, total floor Massing with ~16 min Medium [64]
Office Multiple building, occupied area, programming, area, # courtyards specified floor area
floors courtyard size simulated annealing
Office (M) Spatial configuration: Physically Based Area, adjacency Program layout N/A Med [50]
Multiple semi-automatic methods
floors for layout generation in
practice
Residential (M) Daylight, predicted mean Simulated annealing Programmatic units Aggregation of 4 min Low [65]
Multiple vote, shading modular
floors programmatic units
Residential (M) Maximize area in Rectangular Voronoi Area, weighted Volumetric 12 min Low [66]
Multiple boundary, proximity and Subdivision, Genetic adjacency matrix Arrangement of
floors connectivity of program Algorithm layout
Residential (S) Single Adjacency, size MP 600x400pixel raster Layout on input image 1.3–45.6 s Medium [67]
floor image or vector or vector graphic.
graphic, area,
adjacency
Residential (S) Single Connecting different room N/A Programm graphs, Aggregation of N/A Medium [68]
floor graphs to whole buildings layouts multiple layouts
Residential (M) topology, room dimension Agent based Program graph, area Generated layout ~ Seconds Low [69]
Multiple and aspect ratio, building of rooms assigned to Grid voxel
floors shape
Residential (S) Single Compactness, site GA Areas, adjacency, Program layout 6 s – 7.3 h Medium [70]
floor boundaries, topology, user window door or
rating, circulation, privacy entrance
requirement.
Residential (S) Single None - Exploratory Graph theory Dimensional Program layout ~ 1.5-2 Medium [71]
floor constraints, min
adjacency

Integer Linear Programming [84], Squarified Treemap algorithms [85] education and practice. Distributing design culture through “peer
or more geometry based approaches [22,86]. reviewed” publications of magazines and monographs (a publication
The top-down methods work best when used with fixed boundary describing the body of work of a single architect or architecture office)
constraints. Applied to building design in urban environments, the or through historical or topic specific anthologies and catalogues has
massing of a building is often predetermined (or highly constrained) by analogies to the scientific community. Standardized reference works
local building codes. In a first step, top-down approaches can be used to outlining basic architectural design strategies [111–114] are used for
evaluate whether a certain boundary condition or building massing can teaching the design of building layouts. In both professional and
be filled with a desired program or functional unit. To implement hi­ educational settings, they are used as reference books for dimensioning
erarchies, recursive subdivision methods that iterate over the resulting of standardized building elements, such as stairwells, circulation, esca­
subspaces or programmatic clusters. Working on the end of a hierar­ lators, or bathroom layouts.
chical system, the top-down methods are only able to cover a small, With technological advances in computation and ML, there has been
previously defined design space; in architectural practice that would a renewed interest in referential automatic layout methods. A high-level
mean that stand alone they are less useful for exploratory design stages overview of the referential method is given in Fig. 6 and a comparison of
where the boundary conditions (e.g. building massing) are not yet different methods in literature in Table 4. Several algorithmic methods
defined. for referential design have been used, the most prominent are ML-
algorithms with deep neural networks such as generative adversarial
networks (GANs), as well as mathematical programming methods to
3.4. Referential methods find closest matches.
A series of databases have been ported to be used for generative or
Learning from precedent has a rich tradition in architectural

6
R.E. Weber et al. Automation in Construction 140 (2022) 104385

Fig. 5. Schematic of top-down automatic space layout creation methods. Starting with a definition of boundary conditions, members and objective functions, an
initial subdivision is evaluated by the objective function. Adjustment of the subdivision parameters results in a final structure.

transfer purposes and converted to annotated images or graph struc­ 4. Discussion

tures. For the creation of functional relationships between programs
[63] 120 commercial real estate plans for single family houses were The previous sections summarize the substantial effort that has
encoded as graphs [125]. A Japanese real-estate image databases from already gone into automated space layout generation with existing
the with 5.3 Million images [121] was ported for the use with ML al­ methods borrowing heavily from advanced computational design and
gorithms [122]. To more effectively train neural networks, a series of machine learning approaches. It seems obvious that the real estate sector
residential floorplan datasets were manually collected and annotated by would embrace methods that can provide vital statistics on the
researchers resulting in RPLAN with 80,000 floorplans [117], Rent 3D marketability of a given massing, such as the number of housing units
with 215 floorplans [126] and CubiCasa5K with 5000 floorplans from that can fit or the ratio of rentable to circulation areas. Given that an
Finnish real-estate marketing material [127]. In industry, floorplan automated floorplan algorithm combined with a structural sizing tool
databases enable algorithmic lookup and reuse of floorplan drawings can deliver a set of drawings that can, in principle, go through permit­
from previous work in the development of new layouts [41,52]. ting and be constructed, it seems equally intuitive that many architects
The combination of large image libraries of floorplans with GANs eye such methods with suspicion. The level of detail that such methods
enabled the creation of programmatic infills into arbitrary floorplan provide can create an impression of finality that one traditionally only
shapes for apartment layouts [116] and allowed for the transfer of encounters during later design stages. There is perceived real risk that
different historical architectural styles to apartment floorplans [118]. architects further lose control of the design process at a time when only
Recognizing the importance of hierarchies, strategies such as sequential 2% of US homes are designed by licensed architects. Will that number
infills (starting with the living room as high importance) [84] or addi­ fall even lower?
tional graph networks that inform the generation [120] or training data We find such thinking somewhat defensive. Rather than hanging on
[122,124] highly improve the plausibility of generated floorplans. to the last 2%, should the profession not focus on the lost 98% by creating
Featuring online web interfaces, users can manipulate programmatic the best possible design in the most efficient manner? How can the
graphs while seeing a corresponding architectural layout in real time disciplinary knowledge inform design automation to provide better
[120,123,124]. However, an emphasis is laid on connectivity of rooms quality and more resource efficient spaces and housing?
and their sizes or boundary conditions could not be influenced. As generative methods can produce an infinite range of different
The image-based machine learning methods, however, only work on solutions a variety of heuristics are used to classify promising solutions
very constrained boundaries and small scales, as all information has to or guide optimization processes. This creates an opportunity to include
be encoded in a 256 × 256 pixel image. Even though they can be very building performance as a driver for design generation, extending the
accurate inside of a specific domain and create diverse solutions, purely geometric objectives such as adjacencies, position, or aspect
because of scale limitations, they have only been applied to single story ratio. Validated methods for building energy simulation and natural
residential apartment layouts so far. Furthermore, the fuzzy outputs of ventilation with EnergyPlus [128], and daylight simulation using
image-based ML algorithms require significant postprocessing to Radiance [129] have been integrated into layout automation workflows
recreate usable geometries, while using significant computational power [8,104,130]. Metrics further expanded to include views [110] and agent
and greatly varying in speed. based simulated of human behavior for both characterization and gen­
The strong dependence of the qualities of the outputs on good eration of new floorplans [106,131–133].
datasets, makes the lack of involvement of a diverse representation of A big challenge in the creation of coherent layouts is the problem of
the design community highly problematic. The large-scale datasets used scale. As programmatic requirements get more complex it becomes more
so far in research are based on availability and have not been peer- difficult to coherent layouts that can integrate layers of hierarchy. This
reviewed or curated appropriately for architectural, spatial, or cultural requires either a multi-step approach where programmatic units are
qualities or environmental impact, creating unpredictable outputs. clustered together and subdivided individually [84] or smaller units
(such as a single apartment) are created on their own and then

7
R.E. Weber et al. Automation in Construction 140 (2022) 104385

Table 3
Comparison of top-down automatic design methods in literature.
Typology Scale Objective Function Optimizer Inputs Output Speed Architectural Citation
Quality

Office (S) Single Adjacency, minimize Quadratic Areas, adjacency Assigned program on High Very Low [87]
floor travel distance assignment existing layout
Office (M) Adjacency GA Areas, adjacency Assigned program on High Very Low [88]
Multiple existing layout
floors
Office (M) Adjacency GA Areas, adjacency Assigned program on High Very Low [89]
Multiple existing layout
floors
Office (M) Adjacency GA Areas, adjacency Assigned program on High Very Low [90]
Multiple existing layout
floors
Hospital (M) Adjacency GA Areas, adjacency Assigned program on N/A Low [91]
Multiple existing layout
floors
Residential (S) Single Adjacency, room size MP Adjacency, area, min Assigned program on N/A Medium [92]
floor width/depth, existing layout
Residential (S) Single Adjacency, room size GA, MP Areas, adjacency Design topology (with N/A Medium [93]
floor adjacencies) (tree) and
assigned program on
existing layout
Residential (S) Single Adjacency, heating GA Program description Layout in bounding box 188 s Medium [94]
floor cost, lighting cost, (with min and max
spatial efficiency size), bounding box
Residential (S) Single Custom fitness GA Areas, adjacency, Assigned program on 600 s Low [95]
floor proportions, building existing layout (multiples
perimeter of square foot units)
Residential (M) Adjacency Stochastic Search Adjacency, perimeter Assigned program on N/A Low [96]
Multiple existing layout
floors
Residential (M) Adjacency GA Connectivity, area, Assigned program on N/A Low [97]
Multiple ratio existing layout
floors
Residential (S) Single Practicality, originality, GA, NSGA-II Areas Assigned program on N/A Low [98]
floor user input existing layout
Residential (S) Single Aspect Ratio, area GA Area Assigned program on N/A Low [99]
floor existing layout
Residential (S) Single Areas Squarified Areas, connectivity Assigned program on N/A High [85]
floor Treemap existing layout
KD Tree
Residential (S) Single Areas, connectivity GA Connectivity Assigned program on N/A Low [100]
floor existing layout
Residential (S) Single Areas, connectivity GA Connectivity, Assigned program on N/A Low [101]
floor hierarchy existing layout
Residential (M) Areas, Connectivity, Non-linear least Connectivity, Areas, Rooms inside boundary, 2–3.5 s Med [102]
Multiple Material constraints squares Wall fabrication precast concrete walls Seconds
floors specification
Residential (S) Single Areas, connectivity GA Connectivity, Assigned program on N/A Low [103]
floor hierarchy existing layout
Residential (M) Adjacency, thermal MP Areas, connectivity Assigned program on N/A High [104]
Multiple performance existing layout
floors
Office (M) Gap spaces MP Room templates Rooms tiles in existing Minutes Med [105]
Multiple grid ~80s
floors
Trade Fair (S) Single Mobility, accessibility Stochastic, Agent behavior Rooms inside boundary 2–7 High [106]
floor and coziness of agent- Simulated Minutes
based crowd simulation annealing
Office (S) Single Compactness GA Program description, Room tiles in existing ~520 s Low [107]
floor ~47 geometric grid Minutes
properties
Hospitals (S) Single Fitted program, view, K-D Tree, Program description, Rooms inside boundary N/A Med [19]
floor travel distance, Human
proportion evaluation
Trade Fair (S) Single Congestion, exposure GA, NSGA-II Boundary, program Program distributed in 5 days Med [108]
floor description boundary 20s per
iteration
Residential, (S) Single Gap area MIQP Site boundary, Rooms inside boundary ~15 s High [84]
office floor program description, Seconds
Generic (S) Single Orientation, adjacency, Optimizer (N/A) Site boundary, Rooms inside boundary N/A High [109]
floor user selected + Reinforce- program description
subdivision grammar ment learning
Generic (S) Single Visibility, Tree Depth, Covariance Parametrized Optimized wall layout 2.25–7.41 s Med [110]
floor Entropy Matrix geometric model Seconds
Adaptation

8
R.E. Weber et al. Automation in Construction 140 (2022) 104385

Abbreviations: Genetic Algorithms (GA). Mathematical programming (MP).

Fig. 6. Schematic of referential automatic space layout creation methods. Starting with a dataset (catalogue) of existing structures. A desired property is extracted
from the dataset and a model prepared and trained as Neural Net, TSP or referential data cluster. The model is applied to a user specified problem formulation and
postprocessed into geometric data resulting in a final structure.

Table 4
Comparison of referential automatic layout design methods.
Typology Scale Database Reference Matching Inputs Output Speed Architectural Citation
Source (D) Direct Quality
(P) Post Processed

Residential (S) Single 101 single 256 × 256 pix2pix NN via Boundary Rooms color coded N/A Low [116]
Floor story houses pixel image, RunwayML in boundary (D),
[115] Color Coded manual tracing for
vectors (P)
Residential (S) Single RPLAN [117] 256 × 256 1. CNN for Entrance, Wall map (D), 4 s (Seconds) Medium [117]
Floor pixel image, location of apartment vector (Generation)
Color Coded program 2. CNN boundary representation of 7 Days (ML
for placement of layout (P) Training)
walls
Commercial (S) Single 700 annotated ? x? px pix2pix NN Boundary of Rooms color coded N/A Low [118]
Residential Floor plans (Boston, Image, color building in boundary (D),
Industrial USA, collected coded manual tracing for
by author) vectors (P)
Residential (S) Single 500 floorplans, Program Bayesian model, Apartment Program graph N/A (No [119]
Floor undisclosed graph scored adjacency type architectural
graph representation)
Residential (S) Single RPLAN [117] 128 × 128 1. GNN, CNN Entrance, 128 × 128 0.4 s Medium [120]
Floor pixel image, program Apartment floorplan image (Seconds)
color coded distribution, 2. Boundary, (D), vectorized (Generation)
CNN floorplan Number/Type floorplan (P)
image of rooms
Residential (S) Single 117,587 256 × 256 Conv-MPN Bubble Room masks (D), N/A Medium [122]
Floor Layouts from pixel image, diagram, fitted rectangles as
Lifull [121] program (Program rooms (P)
graph graph)
Office (M) 120,000 Voxel graph, 1. GNNs for the Program Volumetric pixel N/A Low [123]
Multiple volumetric program pro- gram graph Graph, User grid representation (Generation)
Floors designs (by graph 2. GNN voxel input during of program 20 min (with
authors) graph, generation. user
interaction)
Residential (S) Single RPLAN [117] RPlan 1. Relational Program graph Vector <0.4 s ~ High [124]
Floor images GAN, 2. Conv- representation of Realtime
parsed as MPN layout (Generation)
program
graph

9
R.E. Weber et al. Automation in Construction 140 (2022) 104385

assembled as units into a larger buildings [130]. Hierarchical ap­ Table 5

proaches have also been successfully implemented to inform ML models, A new set of holistic metrics to guide automated building layouts.
where placing the living room first in the creation of apartment floor­ Spatial Environmental Structural
plans increased the quality of solutions [117].
Modularity Daylight Spans
To support the creation of new hybrid methods, it is important that Minimal change of the Provide access to daylight Building layouts that
spatial, environmental, and structural considerations can work in par­ floorplan necessary to throughout the building, work with minimal spans
allel and inform one another. We propose to expand the list of existing create different while minimizing direct to reduce amounts of
metrics to create workflows that can enable floorplan layouts supporting configurations while solar radiation and glare. structural materials
retaining same overall needed.
the creation of sustainable and high performing buildings (. layout.
Table 5). These metrics can be tested at various scales from indi­ Compactness Ventilation Continuity
vidual room to apartment, floor, or whole buildings for performance Reduction of circulation to Layouts that promote Layouts that stack
testing and optimization. fit more in a building, natural ventilation (cross loadbearing walls and
while minimizing unused ventilation). enable optimal placement
In addition to combining traditional floorplan generators with the
space. of shear walls to enable
above-mentioned performance workflows, we see three specific use continuous carrying of
cases where automated floorplan methods can enrich the current design loads.
process. Adaptability Energy Material Integration
First, for typical urban infills, arguably the most sustainable and Creating layouts that Minimization of building Enable layouts that
enable flexibility of use by energy use by positioning promote structural
urgently needed building typology to accommodate a growing popula­
the inhabitants, creating and layering of less material systems with low
tion, top-down methods provide a natural starting point since many rooms that can be used for conditioned zones such as embodied carbon and
massing parameters have already been set through zoning and setback different functions or circulation to act as buffers integrate fabrication
requirements as well as clients’ desire to maximize buildable area. layouts that enable to the conditioned spaces. constraints such as
different uses at the same prefabricated timber
There, a hybrid approach seems very promising, combining both top-
time e.g. through shielding modules.
down and bottom-up methods to negotiate between programmatic re­ of noise.
quirements and the urban context [130,134]. Referential methods be
used to augment currently prevalent metrics to evaluate layout designs,
such as daylight access, aspect ratios, or material quantities, verifying knowledge. In our survey, we can show how floorplan layout automa­
the design quality or offer alternative spatial layouts. tion is a dynamic field, both in terms of industry developing new tools,
In the case of greenfield developments, bottom-up methods can be and business cases, as well as in academic research. We can see different
useful for quick design exploration by creating topologically different disciplines engaging with the topic, ranging from architectural design
iterations. Material and construction constraints such as bay sizes, research, civil engineering, building physics and technology, as well as
desired spans or prefabricated small-scale units can be integrated into computer graphics.
the members to ensure solutions are feasible. Varying in resolution, the Showing the opportunities of hybrid approaches that go beyond
members of a bottom-up method do not have to be defined as single purely spatial properties (e.g. proportions, areas or connectivity) to
rooms but could be larger units or building parts, that can be refined or create believable floorplans, we see potential to further evaluate layouts
filled using top-down or referential approaches. based on environmental, and structural constraints that can serve the
A third use case relates to building stock analysis. By applying occupants. We propose the hybridization of the three methods, coupled
floorplan generator to whole neighborhood massing models, existing with a new set of interdisciplinary metrics and performance indicators to
urban analysis methods from daylighting to operational and embodied guide future building layout automation. Working together in an itera­
energy can be significantly refined since a floorplan help quantify the tive loop, the strengths of the different strategies can be applied at
amount of material in a building, the likely number of occupants, and different points in the design process.
the location of internal walls that block daylight. We show how automating building layouts can have a wide range of
value propositions. Current use cases in the real estate industry can be
5. Conclusions expanded to create design tools that utilize automated floorplan layouts
to give feedback about program, occupancy, or embodied carbon in the
In this review, we have surveyed existing automatic floorplan layout early stages of design. Using algorithmic and data driven solutions, they
creation methods in architectural design and have introduced a cate­ can optimize building layouts during the design process or explore
gorization into three methodologies. The bottom-up method proposes to creative solutions for new construction. Furthermore, they could lead to
work with a set of parts, such as rooms or preassembled units and to developing a better understanding of existing building stock or changes
aggregate them into a larger structure. As an exploratory tool, it allows in building policy: empowering architects, urban designers, law makers
for the fast generation of different design options. Aggregation strategies and the public to make more informed decisions towards creating sus­
can be further coupled with heuristics to guide the assembly. However, tainable cities of the future.
navigating often complex constraints or boundary conditions can be
very challenging in the very large design space. There, top-down Declaration of Competing Interest
methods can offer an alternative, starting directly from geometric con­
straints, such as a building or site boundaries that get subdivided into The authors declare that they have no known competing financial
smaller units. For this, different subdivision or packing strategies can be interests or personal relationships that could have appeared to influence
deployed. Third, referential methods are being investigated to make use the work reported in this paper.
of existing buildings and datasets. Geometric properties of existing or
premade layouts can be fit or adapted to a new context. Fueled by recent Acknowledgment
advances in machine learning algorithms, spatial relationships have
been captured as graphs or bitmap images and encoded into neural This research was primarily sponsored by the Sustainable Design Lab
networks, enabling lookup and synthesis. and the Digital Structures group at MIT.
The further accessibility of machine learning algorithms and ad­
vancements of computational tools integrated into traditional geometric
modeling environments used in architectural design could help bridge
the interdisciplinary gap for architects to apply more domain specific

10
R.E. Weber et al. Automation in Construction 140 (2022) 104385

References [33] Autodesk, Project Refinery. https://www.autodesk.com/campaigns/refi

nery-beta, 2020 (accessed October 21, 2020).
[34] M. Dennemark, M. Rudolph, B. Wannemacher, Form Follows You. https://form
[1] M. Wörsdörfer, T. Gül, J. Dulac, T. Abergel, C. Delmastro, P. Janoska, K. Lane,
followsyou.com/typenhausplus/, 2020 (accessed May 13, 2022).
A. Prag, Perspectives for the Clean Energy Transition – The Critical Role of
[35] Kohn Peterson Fox (KPF), Scout. https://scout.build/, 2020 (accessed October 30,
Buildings, Paris. www.iea.org, 2019.
2020).
[2] M.C. Rehm, Complicit: the creation of and collaboration with intelligent
[36] Ramboll, SiteSolve. https://www.site-solve.co.uk/, 2021 (accessed May 16,
machines, Architectural Design 89 (2019) 94–101, https://doi.org/10.1002/
2022).
ad.2417.
[37] A. Andrejevic, J. Spiegeler, B. Worms, omrt - ostate. https://www.omrt.tech/,
[3] M. Del Campo, S. Manninger, A. Carlson, Imaginary plans the potential of 2D to
2021 (accessed May 13, 2022).
2D style transfer in planning processes, in: Ubiquity Auton. - Paper Proceedings of
[38] B. Copley, D. Cairns, citybldr. https://www.citybldr.com/, 2021 (accessed May
the 39th Annual Conference of the Association for Computer Aided Design in
16, 2022).
Architecture ACADIA 412–418, 2019, ISBN 9780578591797 (2019).
[39] A. Hengels, B. Ahmes, Parafin. http://parafin3d.com/, 2021 (accessed May 16,
[4] R.S. Liggett, Automated facilities layout: past, present and future, Automation in
2022).
Construction 9 (2000) 197–215, https://doi.org/10.1016/S0926-5805(99)
[40] A. Pacheco, Gensler launches Blox, an algorithm-powered design visualization
00005-9.
and computation tool, Archinect.Com. https://archinect.com/news/article/150
[5] V. Calixto, G. Celani, A literature review for space planning optimization using an
202814/gensler-launches-blox-an-algorithm-powered-design-visualization-and-
evolutionary algorithm approach: 1992-2014, in: Siggradi, Blucher Design
computation-tool, 2020 (accessed May 13, 2022).
Proceedings, 2015, https://doi.org/10.5151/despro-sigradi2015-110166.
[41] J. van Lith, PlanFinder. www.planfinder.xyz, 2022 (accessed May 16, 2022).
[6] K. Dutta, S. Sarthak, Architectural space planning using evolutionary computing
[42] R.G. Black, S. Duff, A model for teaching structures: finite element analysis in
approaches: a review, Artificial Intelligence Review 36 (2011) 311, https://doi.
architectural education, Journal of Architectural Education 48 (1994) 38–55,
org/10.1007/s10462-011-9217-y.
https://doi.org/10.1080/10464883.1994.10734621.
[7] J. Rhee, R. Krishnamurti, P. Veloso, J. Rhee, R. Krishnamurti, Multi-agent space
[43] T. Van Mele, L. Lachauer, M. Rippmann, P. Block, Geometry-based understanding
planning a literature review (2008-2017), in: J.-H. Lee (Ed.), “Hello, Culture”
of structures, Journal of the International Association for Shell and Spatial
18th International Conference, CAAD Futures 2019 Proceedings, Daejeon,
Structure 53 (2012) 285–295. https://block.arch.ethz.ch/brg/files/2012-jiass-va
Republic of Korea, 2019, ISBN 978-89-89453-05-5.
nmele-lachauer-rippmann-block_1380094579.pdf.
[8] T. Du, M. Turrin, S. Jansen, A. van den Dobbelsteen, J. Fang, Gaps and
[44] E. Ben-Joseph, H. Ishii, J. Underkoffler, B. Piper, L. Yeung, Urban simulation and
requirements for automatic generation of space layouts with optimised energy
the luminous planning table: bridging the gap between the digital and the
performance, Automation in Construction 116 (2020), 103132, https://doi.org/
tangible, Journal of Planning Education and Research 21 (2001) 196–203,
10.1016/j.autcon.2020.103132.
https://doi.org/10.1177/0739456X0102100207.
[9] T. Du, S. Jansen, M. Turrin, A. van den Dobbelsteen, Effects of architectural space
[45] T. Dogan, E. Saratsis, C. Reinhart, The optimization potential of floor-plan
layouts on energy performance: a review, Sustainability. 12 (2020) 1829, https://
typologies in early design energy modeling, in: 14th International Conference of
doi.org/10.3390/su12051829.
IBPSA - Building Simulation 2015, BS 2015, Conference Proceedings, 2015,
[10] Y. Friedman, Flatwriter: choice by computer, Progressive Architecture 03 (1971)
pp. 1853–1860. https://web.mit.edu/sustainabledesignlab/publications/BS2
89–101.
015_FloorPlanOptimisation.pdf.
[11] T. Riley, S. Deyong, M. De Michelis, P. Apraxine, P. Antonelli, T. di Carlo,
[46] R.E. Weber, C. Mueller, C. Reinhart, Generative structural design for embodied
B. Cline, The Changing of the Avant-Garde, The Museum of Modern Art, D.A.P./
carbon estimation, in: Proceedings of the IASS Annual Symposium 2020/21 and
Distributed Art Publishers, New York, 2002, ISBN 0870700049.
the 7th International Conference on Spatial Structures, 2021.
[12] G. Stiny, Introduction to shape and shape grammars, Environment and Planning.
[47] N.C. Brown, V. Jusiega, C.T. Mueller, Implementing data-driven parametric
B, Planning & Design 7 (1980) 343–351, https://doi.org/10.1068/b070343.
building design with a flexible toolbox, Automation in Construction (2020),
[13] G. Stiny, W.J. Mitchell, The Palladian grammar, Environment and Planning. B,
https://doi.org/10.1016/j.autcon.2020.103252 (in press). 103252.
Planning & Design 5 (1978) 5–18, https://doi.org/10.1068/b050005.
[48] O. Tessmann, A. Rossi, Geometry as interface: parametric and combinatorial
[14] H. Koning, J. Eizenberg, The language of the prairie: Frank Lloyd Wright’s prairie
topological interlocking assemblies, Journal of Applied Mechanics 86 (2019)
houses, Environment and Planning. B, Planning & Design 8 (1981) 295–323,
1–13, https://doi.org/10.1115/1.4044606.
https://doi.org/10.1068/b080295.
[49] Y. Zhang, C. Mueller, Shear wall layout optimization for conceptual design of tall
[15] C. Alexander, A city is not a Tree, Parts 1 & 2, Architectural Forum 122, 1965,
buildings, Engineering Structures 140 (2017) 225–240, https://doi.org/10.1016/
ISBN 978-0500275108, pp. 58–62, 58–61.
j.engstruct.2017.02.059.
[16] Esri, ArcGIS CityEngine. https://www.esri.com/en-us/arcgis/products/arcgis-c
[50] L. Helme, C. Derix, Spatial configuration: semi-automatic methods for layout
ityengine/overview, 2022 (accessed May 15, 2022).
generation in practice, The Journal of Space Syntax 5 (1) (2014) 35–49. ISSN:
[17] M. Schwarz, P. Müller, Advanced procedural modeling of architecture, ACM
2044-7507, http://joss.bartlett.ucl.ac.uk/journal/index.php/joss/article/vie
Transactions on Graphics 34 (2015) 1–12, https://doi.org/10.1145/2766956.
w/201/pdf.
[18] A. Heuman, Innovator Spotlight: Part I - “Designing Design Tools”. https://www.
[51] F. Amtsberg, Y. Huang, D.J.M. Marshall, K.M. Gata, C. Mueller, Structural
youtube.com/watch?v=MicmKkkM_3o&t=1375s, 2020 (accessed May 13, 2022).
upcycling : Matching digital and natural geometry, in: Advances in Architectural
[19] S. Das, C. Day, A. Hauck, J. Haymaker, D. Davis, Space plan generator, ACADIA
Geometry, Paris, 2020. http://web.mit.edu/yijiangh/www/papers/AAG2020_St
2016 Posthuman front, in: Data, Designers, and Cognitive Machines, Proceedings
ructural_Upcycling.pdf.
of the 36th Annual Conference of the Association for Computer Aided Design in
[52] O. Green, An introspective approach to apartment design, in: DC I/O 2020,
Architecture 106–115, 2016, ISBN 9780692770955.
Design Computation Ltd., 2020, https://doi.org/10.47330/DCIO.2020.
[20] C. Derix, Mediating spatial phenomena through computational heuristics, in: Life
THHT2676.
Information on Responsive Information and Variations in Architecture. -
[53] L. Wilson, J. Danforth, D. Harvey, N. Licalzi, Quantifying the urban experience:
Proceedings of the 30th Annual Conference of the Association for Computer
establishing criteria for performance based zoning, Simul. Ser. 50 (2018)
Aided Design in Architecture. ACADIA 61–66, 2010, ISBN 9781450734714
237–244, https://doi.org/10.22360/simaud.2018.simaud.031.
(2010).
[54] Clarivate, Web of Science. https://clarivate.com/webofsciencegroup/solutio
[21] E. Finucane, C. Derix, P. Coates, Evolving urban structures using computational
ns/web-of-science/, 2022 (accessed April 11, 2022).
optimisation, in: C Soddu (Ed.), Proceedings of the Generative Arts conference,
[55] Google, Google Scholar. https://scholar.google.com, 2022 (accessed May 15,
Milan, Generative Art, 2006. http://www.generativeart.com/on/cic/paper
2022).
sGA2006/37.htm.
[56] Elsevier, Scopus. https://www.scopus.com/, 2022 (accessed May 15, 2022).
[22] L. Wilson, J. Danforth, C.C. Davila, D. Harvey, How to generate a thousand
[57] B. Martens, Z. Turk, T. Cerovsek, CumInCAD. http://papers.cumincad.org/, 2016
master plans: a framework for computational urban design, in: Proceedings of the
(accessed May 15, 2022).
Symposium on Simulation for Architecture and Urban Design, Society for
[58] G. Staib, A. Dörrhöfer, M. Rosenthal, Components and Systems, Birkhäuser, 2008,
Computer Simulation International, San Diego, CA, USA, 2019, pp. 113–120.
https://doi.org/10.11129/detail.9783034615662.
[23] B. Coorey, Archistar. https://archistar.ai/, 2020 (accessed May 13, 2022).
[59] B.M.A. Rosenman, J.S. Gero, Evolving designs by generating useful complex gene
[24] C. Harness, R. Griege, Test Fit. https://testfit.io/, 2019 (accessed May 13, 2022).
structures, in: P. Bentley (Ed.), Evolutionary Design by Computers, Morgan
[25] R. Gidei, M. Thorley, D. Flynn, Matterlab. https://www.matterlab.co/, 2019
Kaufmann, San Francisco, CA, USA, 1999, pp. 345–364. http://pdf.aminer.
(accessed May 13, 2022).
org/000/743/586/evolving_design_genes_in_space_layout_planning_problems.pd
[26] H. Haukeland, C. Christensen, Spacemaker. https://www.spacemakerai.com/,
f.
2016 (accessed May 13, 2022).
[60] M. Rosenman, Case-based evolutionary design, in: C. Fonlupt, J.-K. Hao,
[27] KREO, Modular. https://www.kreo.net/, 2020 (accessed May 13, 2022).
E. Lutton, M. Schoenauer, E. Ronald (Eds.), Artificial Intelligence for Engineering
[28] S.V. Patel, C. Weijenberg, Digital Blue Foam. https://www.digitalbluefoam.com/,
Design, Analysis and Manufacturing 14, Cambridge University Press, 2000,
2022 (accessed May 13, 2022).
pp. 17–29.
[29] Sidewalklabs, Delve. https://hello.delve.sidewalklabs.com/, 2020 (accessed May
[61] S.A. Arvin, D.H. House, Modeling architectural design objectives in physically
13, 2022).
based space planning, Automation in Construction 11 (2002) 213–225, https://
[30] J. Wallgren, Finch 3d. https://finch3d.com/, 2020 (accessed May 13, 2022).
doi.org/10.1016/S0926-5805(00)00099-6.
[31] B. Wood, Seismic School App. https://seismic-school-app.io/, 2019 (accessed
[62] M. Inoue, H. Takagi, Layout algorithm for an EC-based room layout planning
October 21, 2020).
support system, in: SMCia 2008 : IEEE Conference on Soft Computing in
[32] Mayor of London, Prism App. https://www.prism-app.io/, 2019 (accessed
Industrial Applications, 2008, pp. 165–170, https://doi.org/10.1109/
October 16, 2020).
SMCIA.2008.5045954.

11
R.E. Weber et al. Automation in Construction 140 (2022) 104385

[63] P. Merrell, E. Schkufza, V. Koltun, Computer-generated residential building [92] B. Medjdoub, B. Yannou, Separating topology and geometry in space planning,
layouts, ACM Transactions on Graphics 29 (2010) 1–12, https://doi.org/ CAD Computer-Aided Design 32 (2000) 39–61, https://doi.org/10.1016/S0010-
10.1145/1882261.1866203. 4485(99)00084-6.
[64] F. Bao, D.M. Yan, N.J. Mitra, P. Wonka, Generating and exploring good building [93] J.J. Michalek, R. Choudhary, P.Y. Papalambros, Architectural layout design
layouts, ACM Transactions on Graphics 32 (2013), https://doi.org/10.1145/ optimization, Engineering Optimization 34 (2002) 461–484, https://doi.org/
2461912.2461977. 10.1080/03052150214016.
[65] H. Yi, Y.K. Yi, Performance based architectural design optimization: automated [94] R. Baušys, I. Pankrašovaité, Optimization of architectural layout by the improved
3D space layout using simulated annealing, in: 2014 ASHRAE/IBPSA-USA genetic algorithm, Journal of Civil Engineering and Management 11 (2005)
Building Simulation Conference, 2014, pp. 292–299. http://www.scopus.com/i 13–21, https://doi.org/10.3846/13923730.2005.9636328.
nward/record.url?scp=84938862924&partnerID=8YFLogxK. [95] H. Homayouni, A Genetic Algorithm Approach to Space Layout Planning
[66] I. Chatzikonstantinou, A 3-dimensional architectural layout generation procedure Optimization, 2007, p. 137. http://citeseerx.ist.psu.edu/viewdoc/summary?
for optimization applications, in: DC-RVD, Proceedings of 2014 ECAADe doi=10.1.1.114.7729.
Conference Vol. 1, 2014, pp. 287–296. http://papers.cumincad.org/cgi-bin/wor [96] K. Terzidis, AutoPLAN, Z Joint Study Journal (2007) 84–87. http://ftp.formz.
ks/paper/ecaade2014_163. com/jointstudy/JS2008/11AutoPlan.pdf.
[67] H. Hua, Irregular architectural layout synthesis with graphical inputs, [97] A. Doulgerakis, Genetic Programming + Unfolding Embryology in Automated
Automation in Construction 72 (2016) 388–396, https://doi.org/10.1016/j. Layout Planning, University College London, 2007. http://eprints.ucl.ac.
autcon.2016.09.009. uk/4981/.
[68] B. Dillenburger, Raumindex, Ein datenbasiertes Entwurfsinstrument, ETH Zurich, [98] A. Banerjee, J.C. Quiroz, S.J. Louis, A model of creative design using collaborative
2016, https://doi.org/10.3929/ethz-b-000161426. interactive genetic algorithms, in: Design Computing and Cognition ’08 -
[69] Z. Guo, B. Li, Evolutionary approach for spatial architecture layout design Proceedings of the 3rd International Conference on Design Computing and
enhanced by an agent-based topology finding system, Frontiers of Architectural Cognition, Springer, 2008, pp. 397–416.
Research 6 (2017) 53–62, https://doi.org/10.1016/j.foar.2016.11.003. [99] M.K. Thakur, M. Kumari, M. Das, Architectural layout planning using Genetic
[70] A. Bahrehmand, T. Batard, R. Marques, A. Evans, J. Blat, Optimizing layout using Algorithms, in: Proceedings - 2010 3rd IEEE International Conference on
spatial quality metrics and user preferences, Graphical Models 93 (2017) 25–38, Computer Science and Information Technology. ICCSIT 2010. 4, 2010, pp. 5–11,
https://doi.org/10.1016/j.gmod.2017.08.003. https://doi.org/10.1109/ICCSIT.2010.5565165.
[71] S. Bisht, Transforming an Adjacency Graph into Dimensioned Floorplan 0, 2022, [100] K. Knecht, R. Koenig, Evolutionäre Generierung von Grundriss-Layouts mithilfe
pp. 1–18, https://doi.org/10.1111/cgf.14451. von Unterteilungsalgorithmen, Arbeitspapiere Inform. Der Archit. 10 (2011)
[72] J. Sanchez, Architecture for the Commons: Participatory Systems in the Age of https://e-pub.uni-weimar.de/opus4/files/1653/InfAR_10_Layout_Unterteilung_
Platforms, Routledge, 2020, ISBN 9781138362369. pdfa.pdf.
[73] G. Retsin, Toward discrete architecture: automation takes command, in: Ubiquity [101] R. Koenig, S. Schneider, Hierarchical structuring of layout problems in an
Auton - Paper Proceedings of the 39th Annual Conference of the Association for interactive evolutionary layout system, AIEDAM - Artificial Intelligence for
Computer Aided Design in Architecture ACADIA 2019, 2019, ISBN Engineering Design, Analysis and Manufacturing 26 (2012) 129–142, https://doi.
9780578591797, pp. 532–541. org/10.1017/S0890060412000030.
[74] P. Merrell, D. Manocha, Model synthesis: a general procedural modeling [102] H. Liu, Y.L. Yang, S. Alhalawani, N.J. Mitra, Constraint-aware interior layout
algorithm, IEEE Transactions on Visualization and Computer Graphics 17 (2011) exploration for pre-cast concrete-based buildings, The Visual Computer 29 (2013)
715–728, https://doi.org/10.1109/TVCG.2010.112. 663–673, https://doi.org/10.1007/s00371-013-0825-1.
[75] A. Newgas, Tessera: A Practical System for Extended WaveFunctionCollapse, [103] R. Koenig, K. Knecht, Comparing two evolutionary algorithm based methods for
Association for Computing Machinery, 2021, https://doi.org/10.1145/ layout generation: dense packing versus subdivision, AIEDAM - Artificial
3472538.3472605. Intelligence for Engineering Design, Analysis and Manufacturing 28 (2014)
[76] M. Gumin, Wave Function Collapse. https://github.com/mxgmn/WaveFunctio 285–299, https://doi.org/10.1017/S0890060414000237.
nCollapse, 2021 (accessed May 15, 2022). [104] E. Rodrigues, A. Rodrigues, Á. Gomes, Automated approach for design generation
[77] J. Tóth, J. Pernecký, MONOCEROS. https://monoceros.sub.digital/, 2021 and thermal assessment of alternative floor plans, Energy and Buildings 81 (2014)
(accessed May 16, 2022). 170–181, https://doi.org/10.1016/j.enbuild.2014.06.016.
[78] T. Hosmer, P. Tigas, D. Reeves, Z. He, Spatial assembly with self-play [105] C.H. Peng, Y.L. Yang, P. Wonka, Computing layouts with deformable templates,
reinforcement learning, in: Proceedings of the 40th Annual Conference of the ACM Transactions on Graphics 33 (2014), https://doi.org/10.1145/
Association for Computer Aided Design in Architecture. Distributed Proximities. 2601097.2601164.
ACADIA 2020, 2020, ISBN 9780578952130, pp. 382–393. [106] T. Feng, L.-F. Yu, S. Yeung, K. Yin, K. Zhou, Crowd-driven mid-scale layout
[79] T. Schwinn, A. Menges, Fabrication agency: landesgartenschau exhibition hall, design, ACM Transactions on Graphics 35 (2016) 1–14, https://doi.org/10.1145/
Architectural Design 85 (2015) 92–99, https://doi.org/10.1002/ad.1960. 2897824.2925894.
[80] D. DeFord, M. Duchin, J. Solomon, Recombination: a family of Markov chains for [107] I.G. Dino, An evolutionary approach for 3D architectural space layout design
redistricting, Harvard Data Science Review (2021), https://doi.org/10.1162/ exploration, Automation in Construction 69 (2016) 131–150, https://doi.org/
99608f92.eb30390f. 10.1016/j.autcon.2016.05.020.
[81] N.A. Sherwani, Algorithms for VLSI Physical Design Automation, Springer US, [108] L. Villaggi, D. Zhao, D. Benjamin, Beyond heuristics: novel design space model for
Boston, MA, 1993, https://doi.org/10.1007/978-1-4757-2219-2. generative space planning in architecture, in: ACADIA, Disciplines + Disruption,
[82] P. Pérez-Gosende, J. Mula, M. Díaz-Madroñero, Facility layout planning. An 2017, pp. 436–445. http://papers.cumincad.org/cgi-bin/works/paper/acadia
extended literature review, International Journal of Production Research 59 17_436.
(2021) 3777–3816, https://doi.org/10.1080/00207543.2021.1897176. [109] N. Saha, J. Haymaker, D. Shelden, Space Allocation Techniques (SAT), in:
[83] A. Drira, H. Pierreval, S. Hajri-Gabouj, Facility layout problems: a survey, Annual ACADIA 2020 - Distributed Proximities, 2020, pp. 248–257. http://papers.cumi
Reviews in Control 31 (2007) 255–267, https://doi.org/10.1016/j. ncad.org/cgi-bin/works/paper/acadia20_248.
arcontrol.2007.04.001. [110] G. Berseth, B. Haworth, M. Usman, D. Schaumann, M. Khayatkhoei, M. Kapadia,
[84] W. Wu, L. Fan, L. Liu, P. Wonka, MIQP-based layout design for building interiors, P. Faloutsos, Interactive architectural design with diverse solution exploration,
in: Computer Graphics Forum 37, 2018, pp. 511–521, https://doi.org/10.1111/ IEEE Transactions on Visualization and Computer Graphics 27 (2021) 111–124,
cgf.13380. https://doi.org/10.1109/TVCG.2019.2938961.
[85] F. Marson, S.R. Musse, Automatic real-time generation of floor plans based on [111] E. Neufert, Bauentwurfslehre, Bauwelt-Verlag, 1936, ISBN 978-3-658-34236-4.
Squarified Treemaps algorithm, International Journal of Computer Games [112] T. Jocher, S. Loch, Raumplilot Grundlagen, kraemerverlag, 2012, ISBN 978-3-
Technology 2010 (2010), 624817, https://doi.org/10.1155/2010/624817. 7828-1551-2.
[86] D. Nagy, L. Villaggi, D. Benjamin, Generative urban design: integrating financial [113] O. Heckmann, F. Schneider, E. Zapel, Floor plan manual housing, fifth, rev,
and energy goals for automated neighborhood layout, SIMAUD ’18: Proceedings Birkhäuser (2018), https://doi.org/10.1515/9783035611496.
of the Symposium on Simulation for Architecture and Urban Design 50 (2018) [114] B. Bielefeld, Spaces In Architecture - Areas, Distances, Dimensions, Birkhäuser,
190–197, https://doi.org/10.22360/simaud.2018.simaud.025. Basel, 2019, https://doi.org/10.1515/9783035619706.
[87] R.S. Liggett, W.J. Mitchell, Optimal space planning in practice, Computer Design [115] Zonda, House Plans. https://www.houseplans.com/, 2021 (accessed May 15,
13 (1981) 277–288, https://doi.org/10.1016/0010-4485(81)90317-1. 2022).
[88] J.H. Jo, J.S. Gero, Space layout planning using an evolutionary approach, [116] N. Peters, Enabling Alternative Architectures - Collaborative Frameworks for
Artificial Intelligence in Engineering 12 (1998) 149–162, https://doi.org/ Participatory Design, Harvard. https://www.nathanpeters.us/enabling-alternat
10.1016/S0954-1810(97)00037-X. ive-architectures, 2018.
[89] J.S. Gero, V.A. Kazakov, Evolving design genes in space layout planning [117] W. Wu, X.-M. Fu, R. Tang, Y. Wang, Y.-H. Qi, L. Liu, Data-driven interior plan
problems, Artificial Intelligence in Engineering 12 (1998) 163–176, https://doi. generation for residential buildings, ACM Transactions on Graphics 38 (2019)
org/10.1016/S0954-1810(97)00022-8. 1–12, https://doi.org/10.1145/3355089.3356556.
[90] R. Jagielski, J.S. Gero, A genetic programming approach to the space layout [118] S. Chaillou, A.I. Architecture, Towards a new approach, Harvard (2019), https://
planning problem, CAAD Futures 1997 (1997) 875–884, https://doi.org/ doi.org/10.9783/9781949057027-006.
10.1007/978-94-011-5576-2_67. [119] J. Landes, H. Dissen, H. Fure, S. Chaillou, Architecture as a Graph - A
[91] P.J. Bentley, The revolution of evolution in design: from coffee tables to hospitals, Computational Approach. https://www.academia.edu/42059688/Architecture_a
in: Proc. of Recent Advances in Soft Computing’98, Springer, 1998. http://citesee s_a_Graph_A_Computational_Approach, 2020 (accessed May 15, 2022).
rx.ist.psu.edu/viewdoc/summary?doi=10.1.1.56.909. [120] R. Hu, Z. Huang, Y. Tang, O. van Kaick, H. Zhang, H. Huang, Graph2Plan:
learning floorplan generation from layout graphs, ACM Transactions on Graphics
39 (2020) 1–14, https://doi.org/10.1145/3386569.3392391.

12
R.E. Weber et al. Automation in Construction 140 (2022) 104385

[121] Lifull, Home Dataset. https://www.nii.ac.jp/dsc/idr/lifull, 2015 (accessed [129] G.R. Ward, R. Shakespeare, Rendering with Radiance: The Art and Science of
November 21, 2020). Lighting Visualization, 1998, ISBN 0-9745381-0-8.
[122] N. Nauata, K.-H. Chang, C.-Y. Cheng, G. Mori, Y. Furukawa, House-GAN: [130] E. Rodrigues, M.S. Fernandes, Á. Gomes, A.R. Gaspar, J.J. Costa, Performance-
Relational Generative Adversarial Networks for Graph-constrained House Layout based design of multi-story buildings for a sustainable urban environment: a case
Generation, 2020, https://doi.org/10.48550/arXiv.2003.06988. study, Renewable and Sustainable Energy Reviews 113 (2019), 109243, https://
[123] K.-H. Chang, C.-Y. Cheng, J. Luo, S. Murata, M. Nourbakhsh, Y. Tsuji, Building- doi.org/10.1016/j.rser.2019.109243.
GAN: Graph-Conditioned Architectural Volumetric Design Generation. htt [131] V. Azizi, M. Usman, H. Zhou, P. Faloutsos, M. Kapadia, Graph-based generative
p://arxiv.org/abs/2104.13316, 2021. representation learning of semantically and behaviorally augmented floorplans,
[124] N. Nauata, S. Hosseini, K.-H. Chang, H. Chu, C.-Y. Cheng, Y. Furukawa, House- The Visual Computer (2021), https://doi.org/10.1007/s00371-021-02155-w.
GAN++: Generative Adversarial Layout Refinement Networks. http://arxiv. [132] D. Schaumann, S. Moon, M. Usman, R. Goldstein, S. Breslav, A. Khan,
org/abs/2103.02574, 2021. P. Faloutsos, M. Kapadia, JOIN: an integrated platform for joint simulation of
[125] H. Wood, Essential House Plan Collection: 1500 Best Selling Home Plans, Home occupant-building interactions, Architectural Science Review 63 (2020) 339–350,
Planners, 2007, ISBN 1931131708. https://doi.org/10.1080/00038628.2019.1662767.
[126] C. Liu, A.G. Schwing, K. Kundu, R. Urtasun, S. Fidler, Rent3D: floor-plan priors for [133] D. Nagy, L. Villaggi, J. Stoddart, D. Benjamin, The buzz metric: a graph-based
monocular layout estimation, in: Proceedings of the IEEE Computer Conference method for quantifying productive congestion in generative space planning for
on Computer Vision and Pattern Recognition. 07-12-June, 2015, pp. 3413–3421, architecture, Technol. + Des. 1 (2017) 186–195, https://doi.org/10.1080/
https://doi.org/10.1109/CVPR.2015.7298963. 24751448.2017.1354617.
[127] A. Kalervo, J. Ylioinas, M. Häikiö, A. Karhu, J. Kannala, CubiCasa5K: A Dataset [134] E. Rodrigues, A.R. Amaral, A.R. Gaspar, Á. Gomes, An approach to urban quarter
and an Improved Multi-Task Model for Floorplan Image Analysis, in: Lecture design using building generative design and thermal performance optimization,
Notes in Computer Science 11482 LNCS, 2019, pp. 28–40, https://doi.org/ Energy Procedia 78 (2015) 2899–2904, https://doi.org/10.1016/j.
10.1007/978-3-030-20205-7_3. egypro.2015.11.662.
[128] D.B. Crawley, C.O. Pedersen, L.K. Lawrie, F.C. Winkelmann, Energy plus: energy
simulation program, ASHRAE Journal 42 (2000) 49–56. https://gundog.lbl.gov/
dirpubs/46002.pdf.

13