Physarum Slime Mold: Agent-Based Network Optimization (2D)

Agent Behaviour

Agent count 2000

Sensor angle 30°

Sensor distance 9

Turn speed 0.35 rad

Trail Field

Deposit amount 5.0

Decay rate 8%

Diffusion 55%

Environment

Maze mode (obstacle wall)

Live Readout

Elapsed steps

Food connected

0 / 6

Network efficiency

–

Trail mass

How it works

This is Jeff Jones' agent-based Physarum model — the same sensory-steering algorithm used in slime-mold network-optimization research (and the one Nakagaki's 2000 experiment showed approximating efficient maze/transport routes). Each of thousands of independent agents:

samples the shared pheromone field at three points ahead — left, center and right sensors, offset by the sensor angle at the sensor distance;
turns toward whichever sensor reads the strongest trail (steers straight if center wins, otherwise rotates by the turn speed toward the stronger side, with a small random tie-break — real sensory-based steering, not a random walk);
moves forward and deposits pheromone at its new position.

Every step the whole trail field diffuses (spreads to neighboring cells) and decays (fades), so only trails that are continuously reinforced by traffic survive — a positive-feedback loop that consolidates many wandering paths into a sparse, efficient network. Food sources act as constant pheromone beacons, pulling the network toward them.

Network efficiency is measured, not decorative: every second the trail field is treated as a graph (a cell is traversable once its pheromone exceeds a threshold) and breadth-first search finds the shortest connecting path between each pair of food sources; the ratio of straight-line distance to actual path length is the efficiency shown above — exactly the metric real Physarum-computing papers use to show the organism approximates a low-cost Steiner network. Enable maze mode to add a wall between the food sources and watch the network route around it.

Pheromone trail (agent traffic) Food source (attractant beacon) Maze wall (blocks trail & agents)