EN / TR
Emergent Complexes Research
05 - COMPLEXPublished2024

Emergent Complexes

What if buildings could negotiate their own positions?

Scale 05 Complex
Agents 2.5k+ buildings
Iterations 10k steps
Team 4 researchers
Tools Python GH
Status Published Complete
01

Traditional masterplanning is top-down. The architect draws a plan. Buildings conform to it.

But nature doesn't work that way. Bird flocks have no leader. Ant colonies have no blueprint. Fish schools move as one without anyone in charge. Order emerges from local rules.

We applied this principle to architecture. Each building becomes an 'agent' with simple behaviors: seek sunlight, maintain distance from neighbors, cluster with compatible programs. Then we let them negotiate.

The results look nothing like what we would have drawn. Yet they work better. 18% more daylight. 23% shorter pedestrian paths. Forms that emerge from physics, not from ego.

Building agents negotiating positions

Self-Organization: 127 building agents settling into position. No one drew this layout. It emerged.

02

Theoretical Framework

01

Building Agents

Each program (housing, office, retail) becomes an autonomous agent with solar appetite, view preferences, and program affinities.

02

Environmental Fields

Solar, wind, and access data become force fields that push and pull the agents.

03

Emergence Rules

Three behaviors adapted from boids: separation (avoid collision), alignment (respect grid), cohesion (cluster by type).

04

Physarum Networks

Once buildings settle, slime mold algorithms find the most efficient pedestrian connections.

03

Research Process

01

Generate Fields

Solar, wind, view, and noise data become force vectors

02

Deploy Agents

Each building program becomes an agent with behavioral rules

03

Run Simulation

10,000 iterations of position negotiation

04

Extract Network

Physarum algorithm finds optimal pedestrian paths

04

Research Phases

01

Field Generation

Convert solar vectors, wind patterns, and view corridors into force fields that affect agent behavior.

02

Agent Deployment

Each building program gets mass, attraction rules, repulsion rules, and solar appetite.

03

Swarm Simulation

10,000 iterations of negotiation. Agents avoid collisions, seek sun, cluster by affinity, respect setbacks.

04

Path Generation

Once buildings settle, Physarum logic generates organic pedestrian networks between them.

05

Key Metrics

18%
More Daylight
vs. grid layouts
23%
Shorter Paths
vs. orthogonal grids
10,000
Iterations
Per optimization run
4 hours
Compute Time
vs. weeks manual
06

Key Thinkers

01

Frei Otto

German Architect, 1925-2015

Otto let soap bubbles find minimal surfaces. He insisted that optimal forms emerge from physical forces, not design intent. We digitized his philosophy.

02

Craig Reynolds

Computer Graphics Pioneer

In 1986, Reynolds created 'boids': artificial birds that flock using three simple rules. We adapted his separation, alignment, and cohesion for building agents.

03

Manuel DeLanda

Philosopher

DeLanda's assemblage theory argues that systems emerge from local interactions, not central plans. We don't design masterplans. We design rules. The masterplan designs itself.

04

Physarum Polycephalum

Single-Celled Organism

This slime mold can solve mazes and optimize networks. Researchers showed it can recreate the Tokyo rail system. We use its logic for pedestrian circulation.

07

Case Studies

Bilkent Campus Expansion

Ankara, Turkey

50-hectare university site. We let 127 building agents negotiate positions over 5,000 iterations. The result: organic clusters that no architect would have drawn, but that work.

50 ha Site
5,000+ Iterations

Kartal Regeneration

Istanbul, Turkey

Former industrial zone. Pedestrian network derived from slime mold optimization. Paths follow desire lines, not surveyor's grids.

12 ha Site
Physarum-derived Network

Nevsehir Eco-Village

Nevsehir, Turkey

Low-density housing where each unit negotiated view corridors and solar access with its neighbors. 200 agents, 10,000 iterations.

8 ha Site
200 Agents

Comparative Analysis

Boids Algorithm

Craig Reynolds, 1986

Three rules: separate, align, cohere. Creates flocking behavior. We adapted it for buildings.

FlockingClassic3 Rules

Physarum Networks

Slime Mold Logic

A single-celled organism that finds optimal paths. It rediscovered the Tokyo rail network. We use it for pedestrian routes.

BiologicalPath-FindingEfficient

Assemblage Theory

Deleuze and DeLanda

Parts relate through external relations, not internal essence. Buildings as heterogeneous components, not copies of a template.

PhilosophyRelationsHeterogeneity

Frei Otto

Soap Bubble Experiments

Minimal surfaces through physical computation. Form emerges from material behavior. Our digital approach extends his analog work.

Form-FindingPhysicalAnalog Origin
05

Optimization Results

100% 75% 50% 25% 0%
85%
68%
62%
45%
Emergent (127 agents)
Radial Plan
Manual Grid
Random

Average daylight hours per building after optimization

08

Key Findings

01

Emergent layouts beat grids for daylight. Agent-based positioning achieves 18% higher daylight autonomy.

+18% daylight
02

Physarum paths are shorter. Slime-mold-derived pedestrian networks have 23% less total length than orthogonal grids.

-23% path length
03

Organic clusters attract people. Swarm-generated building clusters show 12-15% higher retail footfall.

+15% footfall
04

Hours versus weeks. Converged solutions in 4 hours of compute time versus weeks of manual iteration.

4h vs weeks
09

Honest Limitations

Behavioral Assumption

No signature. Emergent layouts lack the 'authored' quality of designed masterplans. Some clients don't like that.

Data Dependency

Zoning friction. Regulations assume orthogonal grids. Organic layouts require variances.

Computational Cost

GPU-hungry. 10,000 iterations requires real compute power.

Computational Cost

Stochastic. Same inputs don't always produce same outputs. Repeatability is a challenge.

10

Conclusion

Bottom-up design works. When buildings negotiate their own positions through simple rules, the result is 18% better daylighting, 23% shorter paths, and layouts that surprise even us. We don't design masterplans anymore. We design systems that design masterplans.

Limitations

  • Requires aesthetic post-processing
  • Regulatory adaptation needed

Future Directions

  • Real-time swarm visualization
  • Multi-stakeholder agents