Genetic Design Space Explorer

Design Population

Population size 30

Mutation rate 0.12

Selection pressure 3

Landscape ruggedness 5

Run Statistics

Generation

Best fitness

0.00

Mean fitness

0.00

How it works

Real generative-design software (e.g. topology/shape-optimization tools used in engineering) doesn't search a solution space by hand — it evolves a population of candidate designs across generations using a genetic algorithm:

1. Evaluate: fitness(x,y) = height of the landscape at each candidate's coordinates
2. Select:   tournament of size k picks the fitter of k random candidates
3. Recombine: child = blend(parentA, parentB) with a random mix ratio (BLX-α crossover)
4. Mutate:   child += N(0, σ) · mutationRate, then clamp to the design space
5. Elitism:  the single best design always survives unchanged

The terrain is a multi-peaked fitness landscape — several Gaussian "hills" of different height and width layered together, similar in spirit to an NK fitness-landscape model. Ruggedness sets how many competing local optima exist; a rugged landscape traps low-mutation populations on a good-but-not-best hill, while higher mutation lets the population escape local optima at the cost of more wasted generations.

Population size — more candidates explore more of the landscape per generation, at the cost of visual/compute density.
Mutation rate — the standard deviation of the random step applied to each surviving/recombined candidate; too low stalls on a local hill, too high prevents convergence.
Selection pressure — tournament size; higher values push the population toward the current best more aggressively (less diversity, faster but riskier convergence).
The spheres glide smoothly between generations purely for readability — the underlying algorithm is a discrete generational step.

Best design (elite) Candidate population

⚠ Couldn't load the 3D engineThree.js failed to load from the CDN. Check your connection and reload.