Escolar Documentos
Profissional Documentos
Cultura Documentos
Hoda Homayouni
1. Introduction
Space layout planning is one of the most important and complex parts of any
architectural design process. In order to design a building that responds to most of its
related requirements, an architect (or architectural team) should spend much time and
effort on studying the specific situations of the building and all the relationships that
should exist within the building rooms and between the interior and exterior spaces.
Beside the artistic aspect of architectural design that occurs in almost all parts of
the design process, there is a substantial logical process behind the space layout planning
phase. Architects cannot avoid having numbers of trial and error approach to pass that
step. The combinatorial complexity of most floor plan design problems also makes it
practically impossible to obtain a systematic knowledge of possible solutions using pencil
and paper.
1
Computational Approaches to Space Layout Planning Hoda Homayouni
Topological constraints: constraints that are considered over a couple of spaces, i.e.
adjacency between the rooms, adjacency to perimeter of the building, non-
adjacencies or proximities. The numerical criteria that topologically define
arrangements usually come from site analysis questions and user need questions.
Approach no. 1 has been advocated primarily by Maver (Maver, 1970) who
argues that human creativity is superior to that of machines in solving real –world
problems. This approach is prominently represented by Th’ng and Davies (Th’ng and
2
Computational Approaches to Space Layout Planning Hoda Homayouni
Davies, 1975) whose ideas have been taken up by Gentles and Gardner (Gentles and
Gardner, 1978). Although Cross (Cross, 1977) produces some counterevidence, appraisal
programs enable the designer to increase the number of solutions considered. A tenfold
increase has been experienced (Mayer, 1979). But the solutions are still generated
intuitively and the element of randomness in traditional planning remains in the decision-
making.
Approach no.4 was introduced by Grason (Grason, 1968) who generated plans
as dimensionally feasible embeddings of the duals1 of graphs representing at least the
required adjacencies (edges) between rooms (vertices). According to Steadman
(Steadman 1976), Grason’s program fails for problems of more than five rooms.
After that, Mitchell, Steadman, and Liggett (Mitchell et al, 1976) implemented
an exhaustive method for solving problems stated in terms of dimensional and adjacency
requirements. It includes, however, a final optimization step. To solve an n-room
problem, their program searches a permanent library of topologically distinct
“dissections”, i.e., tight rectangular packings of n nonoverlapping rectangles with
unspecified dimensions. For each dissection matching the adjacency requirements,
combinations of wall dimensions are searched so as to minimize, e.g., construction costs.
Earl (Earl, 1977) has shown that the algorithm which generated the library is not
exhaustive for n>=16 and the optimization method has been criticized by Gero (Gero,
1977). Later, Bloch (Bloch, 1978), Krishnamurti and Row (Krishnamurti and Row,
1978), efficiently generated all dissections with n<=10, and Stiny’s “shape grammars”
(Stiny, 1976) demonstrated their usefulness for such and other generating tasks. The two-
step procedure yields a valuable classification of the solutions according to their
topology. However, the size of the problem independent library limited the
computationally feasible values of n to at most 8.
1
In mathematics, a dual graph of a given planar graph G has a vertex for each plane
region of G, and an edge for each edge joining two neighboring regions. The term "dual"
is used because this property is symmetric, meaning that if G is a dual of H, then H is a
dual of G; in effect, these graphs come in pairs.
3
Computational Approaches to Space Layout Planning Hoda Homayouni
2 Annotated Bibliographies
In this section brief summaries of different programs that have been presented in
the field of space layout planning automation are presented.
4
Computational Approaches to Space Layout Planning Hoda Homayouni
The first system begins with filling out a detailed questionnaire and ends with a
robotic factory producing the unique, responsive house. The major portion of this paper
deals with an explanation of possible techniques used to accomplish this system. The
various techniques can be combined to form house design generating systems:
5
Computational Approaches to Space Layout Planning Hoda Homayouni
TECHNIQUE 8- LET THE USER TRY HIS HAND: Now set the user down to
an interactive graphic display tube and let him try to combine his parts. After each move
by the user, the program evaluates the current arrangement and offers criticism, cost, and
encouragement.
6
Computational Approaches to Space Layout Planning Hoda Homayouni
In this approach, in applying graph theory to floor plan layout, rooms are
pictured as labeled nodes possessing certain attributes, such as intended use, area, and
shape. Adjacencies between rooms are indicated by drawing lines (edges) connecting the
nodes representing those rooms. These notions can be implemented by dealing with the
dual graph of a floor plan which is itself treated as a linear graph. An example of such a
floor plan graph is shown in Figure 4, with black nodes. In the floor plan graph, “edges”
and “nodes” will be called “wall segments” and “corners” respectively.
This special type of dual graph of the floor plan is the design representation to
be used for the class of problems described in this paper. The general idea of its
application is to first set down the four nodes and four edges of the dual graph that
represent the four outside walls of a building. Then nodes and edges are added one by
one to the dual graph in response to design requirements and other considerations until a
completed dual graph is obtained.
The incomplete dual graphs that are produced in the intermediate stages of this
design process present special problems. Since edges can be colored, directed, and
weighted, it is not always clear whether or not there exists at least one physically
realizable floor plan satisfying the relationships expressed in the incomplete dual graph.
To treat this problem, appropriate properties of the dual graph representation have been
developed and are presented in this paper, but there is not sufficient space here to
describe these methods.
7
Computational Approaches to Space Layout Planning Hoda Homayouni
In addition to simply storing the structural description of the dual graph, the
computer data representation must be configured in such a way that it is easy to
determine whether the graph is planar or not. It must also be easy to generate the various
possible geometric realizations of the dual graph. A geometric realization of a planar
graph is simply one of the possibly many ways in which it can be drawn in a plane. Four
different realizations of a particular planar graph are shown in Figure 5.
8
Computational Approaches to Space Layout Planning Hoda Homayouni
To deal with the problems mentioned above, a special graphical grammar called
the planar graph grammar (PGG) was developed. The general intent of this grammar is to
divide a graph into a set of hierarchically organized subgraphs. Each of these subgraphs
can be treated as a separate entity, and if it is connected to the rest of the graph, it is by at
most two nodes. These subgraphs are free to rotate about these nodes to create the various
realizations of the graph. Other such “degrees of freedom” in creating realizations are
also possible.
The grammar allows one to deal with planarity in an inductive manner. That is,
given a graph that is planar and described in terms of the planar graph grammar, if a new
edge is to be added to that graph a test can be developed to tell whether the resultant
graph will also be planer. This test assumes that each of the special subgraphs is already
planar. It then checks to see if the proposed edge can be connected to the two endpoints
without crossing an edge of one of these subgraphs. This method is extremely fast, and its
computation time increases only linearly with the size of the graph considered. The
computation time for other methods generally increases more than linearly with the size
of the graph.
At the next level, the program has two jobs to accomplish: (1) It must satisfy the
physical dimension requirements. It does this by selecting nodes on the boundary of the
region in question and assigning dimensions to them from the design requirement list. (2)
It must “fill” the region with edges representing those adjacencies not specifically
requested by the design requirements. Here the program does an exhaustive search of all
possible design solutions.
The spaces and walls are modeled as point masses, with adjacencies between
spaces that are modeled as springs connecting the masses. Objectives specified in the
architectural program are translated into forces applied to the masses.
In the first phase the topological objectives apply forces to the center of a space.
For collision detection, in this step boundary shapes are treated as circles so spaces are
9
Computational Approaches to Space Layout Planning Hoda Homayouni
able to slide around each other. The dynamic simulation starts to run until the system is in
equilibrium, which is defined as the point in time when the magnitudes of all velocities
are below a small threshold.
At the second phase topological objectives are turned off and geometric
objectives apply forces to the polygonal edges of space boundaries. Also space
boundaries are switched from a circular to a polygonal representation in this phase.
Collision detection and response then act to keep spaces from overlapping, resulting in an
arrangement that is very close to a recognizable building floor plan.
Once a geometric simulation has reached equilibrium, the designer can begin to
analyze and interact with the design by directly manipulating the graphic model rather
than by respecifying design objectives in the language of the underlying system. The
mass-spring representation allows the graphic model to adapt to those changes
immediately. The intention here is not simulating the actual behavior of building
elements, but simulating the way architects may view and interact with design elements
during their conception. Sample pictures for the process are shown in Figure 6.
10
Computational Approaches to Space Layout Planning Hoda Homayouni
Here, two constraint groups are defined in an extensible library: (1) specification
constraints; (2) research space reduction constraints.
Figure 7. Permissible position for (e2.x2, e2y2) after non-overlapping constraint with e1. The
partitioning of the surroundings of a space in {E,W,N,S} is given.
11
Computational Approaches to Space Layout Planning Hoda Homayouni
After the detection of the most constrained space with the building unit, its
corresponding adjacency variables with the building unit are instantiated. After each
space choice, one updates dynamically the dg-cont values of the remaining spaces. The
process runs until all topological variables are instantiated but a backtracking is
performed as soon as an inconsistency is detected. At this point the graphic representation
of a topological solution is generated which reveals slight overlapping of rectangles in the
same way as a sketch.
12
Computational Approaches to Space Layout Planning Hoda Homayouni
The objective functions minimize the wall length or the corridor surface area.
Thus one of the major objectives that remains is to carry out multi-criteria optimization.
The optimal geometrical solution of each topological solution is displayed in a collector
of geometrical solutions. The most important function of this collector is to realize a
classification of the topological solutions from the minimal objective function value of
the geometrical solutions related to each topology. It can be noted that several
geometrical solutions can correspond to the same optimum. In that case, they are all
enumerated.
Figure 9. Some optimal geometrical solutions of the house with two floors.
13
Computational Approaches to Space Layout Planning Hoda Homayouni
3 Conclusion
2. Different techniques and methods are presented in this field where each has
focused on different areas of space layout planning and optimization.
3. Most of these approaches have stopped in the research stage and are not evolved
into the marketing stage.
To get to the point, one of the reasons that these programs are not applicable in
the marketing stage is their excessive runtime. On one hand, the runtime of these
programs increases exponentially by increasing the objectives so that the program can not
handle calculating complexities. On the other hand, the best application of such programs
could be the complex projects. In fact, architects usually have enough experience and
knowledge to solve the design objectives in small projects that they do not need to use
additional helping software. Furthermore, architects are capable to consider aesthetic
aspects of design while this job is not easy for programmers.
Future work would involve finding the best program among the presented
methods which its runtime could become close to a linear function of the number of
objectives. Also, according to the limitations that exist in using different forms for
different spaces, the chosen program should be able to present solutions at the end of
topological stage. This will let the architect to use his tact and aesthetic aspects in the
design.
In fact, an imaginable near future for these approaches, is the situation that the
software could help the architect in the design process especially in analyzing and
optimizing tasks. It is important to highlight that the software should be an assistant to the
architect and not the opposite.
4 References
14
Computational Approaches to Space Layout Planning Hoda Homayouni
Jo JH, Gero JS. Space layout planning using an evolutionary approach. Artif
Intell Engng (1998), 12:149-62.
15
Computational Approaches to Space Layout Planning Hoda Homayouni
Hiller, M., Kolbe, O., Bayer, W., and Ruhrman, I. Heuristic solution of general
allocation problems in administration and regional planning. Bulletin of Computer Aided
Architectural Design 20 (May 1976), 30-36.
16
Computational Approaches to Space Layout Planning Hoda Homayouni
Tsang EPK, Borrett JE, Kwan ACM. An attempt to map the performance of a
range of algorithm and heuristic combinations. Hybrid problems, hybrid solutions. In
Proceedings of AISB. IOS Press (1995). p. 203-16
17
Computational Approaches to Space Layout Planning Hoda Homayouni
18