The document presents Matei Lunca's master's thesis on automatically generating 3D environments. It discusses previous work, demonstrates a software tool for world-scale and object-scale generation, and analyzes parameters, algorithms, results, and techniques like the Variant-Linker template. The goal is parameter-driven world and object generation that combines human design skills and computer speed/calculation.

1. Final Presentation of Thesis
Automatic Generation
Of 3D Environments
INF/SCR-06-53
Thesis student: Matei Lunca (Master Game And Media Technology)
Thesis professor: Dr. R. van Oostrum, GIVE
Thesis supervisor: Drs. A. Kamphuis, GIVE

2. • Presentation Outline
– Why automatic generation of 3D is great
– Some salient previous work
– Demonstration video of software tool
– Parameters
– Algorithm (world-scale generation)
– Results
– The Variant-Linker Template (object-scale generation)

3. • Modelling 3D
– great volume of operations = work = errors
– software should do as much work as possible
• Generator
– contains formulas that create geometry based on
many random numbers and some variables
– variables are user-defined parameters
• “minimum distance between each pair of buildings = 9”
Don’t create,
generate

4. Don’t create,
generate

5. • What is needed for automatic generation of
3D that has not yet been explored?
– parameter-driven world-scale generation
– object-scale generation that uses both the
advantages of human users (overview, “feel” for
architecture” and the advantage of computers:
speed of calculation
Crucial factor

6. • performance improving through testing
– virtual agent prototyping
• obstacles, dead-ends, other agents, interaction,
flocking, goals, motion planning, path finding
• emergency/congestion simulation
• eliminating environment borders
• theoretically infinite level of detail
– adding detail to existing 3D
Applications

7. • urban zone GIS applications
– forecasting patters of urban growth
• minimizing of storage; just-in-time creation
– Xbox 360 procedural synthesis
• rich backgrounds for animate first, model later
• aiding in brainstorming sessions
– visualisations show ideas faster than mathematics
Applications

8. • eliminating need for domain knowledge of 3D
– enabling other disciplines to visualise 3D
environments that depict their ideas
– knowledge and its representation becomes more
important than technical 3D skill
• eliminating map knowledge from gaming
• design-by-sketch
• more time for non-modelling project tasks
Applications

9. • Various Techniques
– generate a city network based on images and
rules with different road patterns
– models built based on aerial photo/laser survey
together with maps (photogrammetry); skeletons
for buildings and roofs can be reconstructed
– "smart buildings” replicate floor plans, each floor
being an instance of a floor prototype: walls, roofs
Tried techniques

10. • Various Techniques
– Worms 3D' seed modules: chunks of land, features and
objects which are joined together randomly
– Hellgate London's template sets of buildings and corridors
• Established techniques: fractal terrain
• Generation of urban zones - conflicting needs!
– realistic (existing cities come into their form through
different processes: habitation, growth, destruction)
– mimicking this process is more of an art, but requires
mathematical precision in programming
Tried techniques

11. Demonstration video

12. • Some drawbacks inherent to 3D modelling
– lack of proper (world) scale/proportions
– architecture appears unrealistic
• clean, repetitive, never enough detail, never enough
computer power...
– lack of a weathered, "used universe"
Drawbacks

13. • Some are specific to generation
– lack of control over what is being created
– users might lack interest/confidence because less
human effort/ thinking involved
– laziness on the part of modellers and its effect on
users; the intellectual process suffers
– establishing parameter values is trial-and-error
– too much data!
Drawbacks

14. • Study of techniques, urban zones determined
choice of parameters and generator functions
Parameters

15. • building_grids
– the width of the overall matrix within for deciding
zone types to build (business, urban, suburb, empty)
• building_grids_2
– the width of the matrix of specific decisions for
individual buildings within units of the overall matrix
• builds
– the number of times the building generator runs
Parameters

16. • chance_streets_grid
– the percentage chance for a raster pattern of
streets
• chance_streets_star
– the percentage chance for a street pattern laid
out from a central point outward to the terrain
edges
• chance_streets_random
– the percentage chance for streets that connect
more random points
Parameters

17. • chance_grid[type] (4x)
– the percentage chance that a matrix cell of overall
building decisions will be of a type: business,
urban, suburb or empty
• grid[grid][building] (20x)
– the percentage chance for buildings of a certain
type (4x) will be built when building city blocks of
a certain type (4x) + the chance that a city blocks
of a certain type will be rotated
Parameters

18. Building types

19. 1. load input parameters
2. calculate values related to parameters, such as
matrix dimensions, borders, buffers and arrays
3. start returning textual output
4. generate terrain: grow the desired number of
terrain features
5. create the water
Algorithm

20. 6. generate streets according to the given chances for
street network patterns
7. start the generation of buildings by dividing the given
central part of the terrain into a grid
8. divide each segment of this overall grid into more
refined grids
9. decide for each overall grid segment what type of
urban zone it will contain
10. loop through the overall grid, then within that loop
through the local grid, creating building models
according to the chances and desired percentages for
types given as parameters
Algorithm

21. 11.add trees
12.remove the water and street obstacles
13.store environment as VRML or XML
– the first is faster and more storage efficient, the
latter is both machine and human readable
11.take a screenshot
12.report completion and visualise
Algorithm

22. Algorithm

23. Algorithm

24. • Characteristics
– each environment is unlike others
– comparable parameters = comparable output
– many types of zones are created
• areas that cannot be navigated (water)
• varying terrain
• areas with easy "tree" obstacles
• areas with moderate "building" obstacles
• areas with "building” maze: dead ends, narrow
passages, alleys, streets, trees, visual restrictions
Results

25. Results

26. • Variant: describes minimum and maximum volumetric values
• Linker: glues Variants
Variant-Linker

27. Variant-Linker

28. • Advantages
– modularity, and therefore reusability
– validity guarantees through the use of already "proven"
templates, not unlike extensions in programming
– high degree of usability and enabling modelling
– simplicity
– scalability in the way of embedded information
• Disadvantages
– dependency on some human input
– lack of visual attractive output?
– back to speed of normal modelling
Variant-Linker