A slime mold draws the SBB.
Sixteen thousand GPU agents start scattered across a region. Each one drops a faint chemoattractant and follows the strongest gradient it can sense. The cities are constant sources of attractant. Over a minute or two, a transport network emerges between them. The same mechanism that, in vivo, made Physarum polycephalum redraw the Tokyo metro in Tero et al. (2010).
WebGL, 16'384 agents on ping-pong half-float textures · prefers-reduced-motion respected
What you see
Each agent is a single GPU pixel that holds a position and a heading. Forward, left and right sensors check the chemoattractant gradient and the agent turns toward the strongest. A faint deposit at each step makes the trail self-reinforcing.
Why cities
Real cities act as constant chemoattractant sources. The simulation does not know about roads. It only knows that agents are drawn to attractant, and that paths between strong attractants get reinforced. The network emerges from rule plus repetition, not from design.
How fast
About thirty seconds for the broad shape, two minutes for stable connections. Faster in Tokyo (smaller bounded region, more cities close together), slower in Switzerland (cities spread out, agents need to find each other).
Who else has done this
Andrew Adamatzky has run Physarum on physical maps of the UK and Germany since 2010. The Japanese team around Atsushi Tero used a real slime mold on a wet plate to reproduce the Tokyo metro. Sage Jenson's 2018 GPU implementation made the pattern viral in creative coding.
- Jones (2010). Characteristics of pattern formation and evolution in approximations of Physarum transport networks. Artificial Life.
- Tero, Takagi, Kobayashi, Yumiki, Bebber, Fricker, Mahadevan, Nakagaki (2010). Rules for biologically inspired adaptive network design. Science.
- Jenson (2018). Physarum. sagejenson.com/physarum.
- SBB Open Data (data.sbb.ch). Schweizerische Bundesbahnen, Liniennetz GeoJSON.