SLIDE 1
Our Goal
- Catch as much sun as possible!
Growing Trees with the Genetic Algorithm Our Goal Catch as much - - PowerPoint PPT Presentation
Growing Trees with the Genetic Algorithm Our Goal Catch as much - - PowerPoint PPT Presentation
Growing Trees with the Genetic Algorithm Our Goal Catch as much sun as possible! Implementation - Setup Unreal Engine 4 Ray traces Simple fitness function Building of graphical interface The Genetic Algorithm
SLIDE 2
SLIDE 3
Implementation - Setup
- Unreal Engine 4
- Ray traces
- Simple fitness function
- Building of graphical interface
SLIDE 4
The Genetic Algorithm
- General idea - Evolution
- Fitness
- Stochastic selection
- Combining DNA
- Incremental improvement
- Complexity vs Creativity
- Why is GA suitable for our
problem?
SLIDE 5
Implementation - Algorithm
- A functioning algorithm
- Physical and DNA
representation of tree, branches, leafs
- Mutation
- Sexual vs Asexual
reproduction
- Fitness functions
- Convergence
- Population
SLIDE 6
Fitness Function
- A function that evaluates a tree for each generation tick
- Mimics the sun
- Different types
- Experimenting
SLIDE 7
Fitness Function - Improvements
- Parallell rays
- Player controlled functions
SLIDE 8
Fitness Function type - Normal (straight above)
SLIDE 9
Fitness Function type - Manual
- Any direction
SLIDE 10
Fitness Function type - Sweep
- Shoots rays from multiple angles
- Gave somewhat vague results
SLIDE 11
Fitness Function type - Hemisphere
- Trail and error
- Gave good results with increased res.
SLIDE 12
Fitness Straight Above
Generation 1 Generation 10 000
SLIDE 13
Hemisphere Fitness
Generation 1 Generation 16 000
SLIDE 14
Changing Environment
- User controlled obstacles
- Cubes
- Rocks
- Plates
- All scalable and rotatable
SLIDE 15
Comparison - Fitness Straight Above
SLIDE 16
Results
- With what can we compare our results?
SLIDE 17
Implementation - Improving the algorithm
- Sexual reproduction
- Modular data structure for branches
- Soft random selection
- Replacements per generation
- Lower mutation frequency,
more possible mutations
SLIDE 18
Hill climbing
- Should only be performed when GA seems to have converged.
- Reaches local maximum.
- Destroys possibility to continue
genetic algorithm.
SLIDE 19
GUI
SLIDE 20
Conclusion
- Problems, solutions, lessons learned
○ Selection ○ Reproduction ○ Data structure ○ Fitness
- Weaknesses and strengths of GA
○ Creativity ○ Complexity ○ Dependent on ad-hoc algorithms.
- Overall, satisfying results and our goals were reached.