standard tier · 72 briefs · 1,368 runs · results db2e24e

CityFormGen

Generative and optimisation methods for middle housing, compared under matched compute

RESEARCH GENERATIVE-DESIGN BENCHMARK - NOT CONSTRUCTION DOCUMENTATION OR CODE CERTIFICATION.

medium · flats · 2 storeys

B0006: a adu pair on a 12.8 × 22.2 m lot

2 dwellings, 12 rooms. The pooled front across all methods holds 441 nondominated designs.

Requirements

Lot
12.8 × 22.2 m (284 m²)
Street side faces
195° (0 = north)
Setbacks front / rear / left / right
4.2 / 7.4 / 1.3 / 1.3 m
Storeys × height
2 × 2.8 m
Shared core
yes (access gallery + stair)
Open space, net of parking
38% of the lot
Parking spaces
0
Courtyard
not required
Tightness (programme / buildable volume)
0.584

Programme

DwellingTypeStoreyMin m²Rooms
U03BR084entry, living, dining, kitchen, bathroom, bedroom primary, bedroom secondary, bedroom secondary
U1studio133entry, living, kitchen, bathroom

Required contacts: entry–living, kitchen–living. Doors forbidden between: entry–bedroom_primary, entry–bedroom_secondary, kitchen–bathroom.

Results by method

MethodSolvedHVEvalssNotes
Heuristic0%0.0003450.6
Random search0%0.00015002.6
Simulated annealing67%0.07415002.5
NSGA-II67%0.07615003.3
CP-SAT100%0.146476.8
Random sketch + repair33%0.04415003.7repair collapse
Heuristic-seeded NSGA-II100%0.12815005.0

Mean over 1 seeds (the heuristic is deterministic: one run).

Representative designs from the pooled front

Each is best on one objective, or the knee: the most balanced trade-off. None is best overall.

Closest attempts that failed

For methods that found nothing feasible: their least-infeasible design, with its violations marked on its page.

Pareto explorer

Loading designs…

Convergence on this brief

Archive hypervolume against evaluations, mean over seeds.

Heuristic
00.161500 ev
Random search
00.161500 ev
Simulated annealing
00.161500 ev
NSGA-II
00.161500 ev
CP-SAT
00.161500 ev
Random sketch + repair
00.161500 ev
Heuristic-seeded NSGA-II
00.161500 ev

What stopped designs here

Share of each method's evaluated designs that violated each constraint.
methodC04C05C06C07C08C09C10C11
Heuristic8%37%7%5%1%13%100%
Random search95%100%80%99%13%71%73%83%
Simulated annealing62%91%56%63%10%42%41%48%
NSGA-II44%77%39%50%10%33%31%56%
Random sketch + repair62%68%62%20%11%25%58%47%
Heuristic-seeded NSGA-II32%49%23%16%2%22%26%25%

Constraint definitions