Anda di halaman 1dari 34

Computer-Generated Residential Building Layouts

Paul Merrell Eric Schkufza Vladlen Koltun


Stanford University

Modeling Buildings with Interiors


n

Goal: Model the internal structure of buildings Crucial in many interactive applications
q

Buildings that can be entered and explored

Commonly created by hand

Residential Buildings
n

Focus on residential buildings


q

Common in games, virtual worlds Have complex structure

Office buildings and schools


q

Highly regular layouts

Related Work
n

Automated Spatial Allocation


q q

March and Steadman, 1971 Shaviv, 1987

Physically Based Modeling


q q q

Arvin and House, 2002 Mass-spring system Sensitive to initial conditions

VLSI Layout
q

Sarrafzadeh and Lee, 1993


4

Computer Graphics Research

Mller et al., 2006

Whiting et al., 2009

Legakis et al., 2006

Pottmann et al., 2007

Architectural Design in the Real World

Clients high-level specifications

Set of floor plans

- Number of bedrooms - Bathrooms - Total square footage, etc.

Architectural program
Rooms & adjacencies

Exterior style

Overview

Clients high-level specifications

Architectural program
Rooms & adjacencies

Set of floor plans

3D model

First end-to-end approach to automated generation of building layouts from high-level requirements

Possible Approaches to Building Layout Design


n

Use a grammar
q q

Shape grammar [Stiny, 2006] Hard to capture the functional relationships Too many rules of thumb, ill-specified Infer design principles using machine learning techniques

Use guidelines from architects


q

Use a data-driven approach


q

Data-Driven Architectural Programming


n

Sample from a distribution of architectural programs Conditioned on the high-level contraints

Bayesian Network
n

Represent the distribution in a Bayesian network


q

Compact representation

n n

Nodes probabilities Edges conditional dependencies Sample from conditional distributions


q

Bayesian network

Use high level specifications


10

Structure Learning Results


Architectural programs

10 iterations

100 iterations

1,000 iterations

Output one sample

11

Overview

Clients high-level specifications

Architectural program
Rooms & adjacencies

Set of floor plans

3D model

12

Floor Plan Optimization


n

Metropolis algorithm
q q q

Propose a new floor plan Evaluate it, then accept or reject it Not a greedy algorithm

13

Metropolis Algorithm
n

Objective function
Building layout Constant Cost function

n n

In each iteration, propose a new building layout Accept with probability

14

Proposal Moves
n

Slide a wall

Slide the entire wall

Snap walls together


15

Split into two collinear walls

Proposal Moves
n

Swap two rooms

Helps to explore the space more rapidly


16

The Cost Function


n

Evaluates the quality of the layout

Accessibility term

Dimension term

Floor compatibility term

Shape term

17

Accessibility Term
n n

Architectural program specifies adjacencies Outdoor access for entrances, patios, and garage.

Accessibility term excluded


18

Dimension Term
n

Likelihood of a rooms area and aspect ratio


q

Uses Bayesian network

Area term excluded

Aspect ratio term excluded


19

Shape Term
n

Measure concavity of a shape, S

H(S) - S H(S)

20

Shape Term

Shape term excluded

21

Cost Function
n

All terms included

22

Floor Compatibility Term


n

Each floor should be supported by the floor below it

23

Floor Plan Optimization

200 iterations

2,000 iterations

20,000 iterations

100,000 iterations

24

Overview

Clients high-level specifications

Architectural program
Rooms & adjacencies

Set of floor plans

3D model

25

Different Exterior Styles

Cottage

Italianate

Tudor

Craftsman

26

Results

27

Results

28

Results

29

Results

30

Results

31

Future Directions
n n n n n

Non-rectilinear / curved wall segments Site-specific and client-specific factors Integrate structural stability Interactive exploration of layout designs Other building types

32

Conclusion
n

First end-to-end approach to automated generation of building layouts from high-level requirements Data-driven approach to procedural modeling

33

Questions?

34