Watch ants find food using pheromone communication — an emergent self-organised behaviour where simple rules create complex colony intelligence. Each ant follows local chemical gradients, yet the colony as a whole finds optimal paths.
Stigmergy — indirect communication via environment. Ants deposit pheromones that attract others, creating positive feedback loops for efficient foraging.
Use sliders to change ant count, food sources and pheromone decay rate. Click on the field to add food. Right-click to remove it. Toggle two-colony mode to watch competition.
Real ant colonies use over 20 different pheromone chemicals. This model simulates the two most important: "food found" and "way home" trails.
This simulation models ant foraging behaviour using stigmergy — a form of indirect communication where ants deposit chemical pheromone trails that guide other colony members. Each ant follows simple local rules: wander, sense pheromone concentration ahead, steer toward the strongest signal, and deposit your own trail. From these minimal behaviours, the colony as a whole discovers efficient routes to food sources, a classic example of swarm intelligence and emergent complexity.
Ant colony foraging inspired the Ant Colony Optimisation (ACO) family of algorithms, first formalised by Marco Dorigo in the early 1990s. ACO is now widely used to solve combinatorial problems such as vehicle routing, network design, and the travelling salesman problem.
Stigmergy is a mechanism of indirect coordination in which individuals modify their shared environment and others respond to those modifications. In ant colonies, pheromone trails left on the ground act as the shared medium — no ant needs to communicate directly with another. This principle allows colonies of thousands of individuals to collectively solve complex foraging problems without any central controller.
Use the sliders in the panel to adjust the number of ants (10–300), food sources (1–8), pheromone decay rate, ant speed, and sensing view angle. Left-click anywhere on the canvas to place a new food source; right-click to remove one. Toggle "Two colonies" to watch two competing nests with separate pheromone systems. Hit Restart at any time to reset the simulation with the current settings.
Orange/red trails indicate the "to food" pheromone deposited by ants carrying food back toward the nest — it guides searching ants toward discovered food sources. Blue trails represent the "home" pheromone deposited by searching ants, which helps food-laden ants find the nest. The combination of these two overlapping gradient fields is what steers the entire colony toward efficiency.
Pheromone concentration on each grid cell is multiplied by a decay factor (0 < d < 1) every simulation frame, modelling evaporation. The amount remaining after t frames follows an exponential curve: P(t) = P(0) * d^t. When d is close to 1 (e.g. 0.999), trails persist for many seconds; when d is lower (e.g. 0.9), they fade within a fraction of a second. The balance between deposition rate and decay rate determines whether positive feedback can sustain a stable trail.
In the classic double-bridge experiment by Goss et al. (1989), a colony was offered two routes of different lengths between nest and food. Ants randomly chose both initially, but because the shorter route took less time, returning ants reinforced it with more pheromone per unit time before the longer route's trails grew equally strong. The colony reliably converged on the shorter path — the same feedback mechanism this simulation demonstrates with multiple food sources at varying distances.
Correct — this is one of the most important points about swarm intelligence. No ant knows the colony's goal; each follows only local chemical signals and a few simple behavioural rules. The apparently intelligent colony-level behaviour — finding short paths, prioritising richer food sources, recovering from blockages — arises entirely from the interaction of many simple agents. This decentralised self-organisation is the defining feature of complex adaptive systems.
Marco Dorigo introduced the Ant System algorithm in his 1992 PhD thesis at Politecnico di Milano, inspired by ethological studies of foraging in species such as Lasius niger. The first ACO paper was published with Alberto Colorni and Vittorio Maniezzo in 1991. Dorigo later extended it into the Ant Colony System (1997) and MAX-MIN Ant System variants, which remain competitive on benchmark routing problems today. He received the IEEE Frank Rosenblatt Award in 2015 partly for this work.
Boids (flocking) shares the swarm-intelligence foundation: agents follow local rules and global order emerges. Langton's Ant shows a single-agent cellular automaton with surprisingly complex long-term behaviour. The Travelling Salesman Problem simulation on this site uses ACO directly as its solver. Genetic algorithms share the broader class of nature-inspired metaheuristics. Slime mould simulations model a similar deposit-and-follow pheromone mechanism in the organism Physarum polycephalum.
Ant Colony Optimisation is applied in logistics (UPS and FedEx-style routing), network packet routing (AntNet), robot path-planning, protein-folding prediction, and job-shop scheduling. Telecom companies have used ACO-inspired algorithms to dynamically reroute traffic around congestion. In chip design, ACO helps place and route integrated circuit components. Its key advantage is robustness to dynamic changes: because ants keep exploring, the colony can adapt when the environment shifts — an advantage over static optimisers.
Active research areas include hybrid ACO-machine learning systems where pheromone weights are informed by neural network predictions, multi-objective ACO for simultaneous optimisation of cost and reliability, and quantum-inspired variants that explore solution spaces more efficiently. On the biological side, researchers are using calcium imaging and high-speed video to map individual pheromone decision-making in Camponotus and Pogonomyrmex species at millisecond resolution, with findings feeding back into more accurate algorithmic models.
This is an agent-based model of ant foraging driven entirely by local rules. Each ant wanders, senses pheromone concentrations in three directions ahead, and steers toward the strongest signal while depositing its own trail. Two scalar fields are tracked on a grid: a "to food" trail laid by searching ants and a "home" trail laid by returning ones. Decaying trails plus positive feedback let the colony converge on short routes — an example of stigmergy, the principle behind ant colony optimisation.
Emergent path-finding from simple agents. Ants read pheromone in a forward, left and right cone set by the view angle, turn toward the densest cell sampled 12 pixels ahead, and add a small random wander. Trails are stored as two Float32Array grids (4 px cells) that are multiplied by the decay factor every frame, so unused paths fade while reinforced ones persist.
Sliders set ant count (10–300), number of food sources (1–8), pheromone decay (0.9–0.999), ant speed and the sensing view angle (10–120°). Left-click adds a food source; right-click removes a nearby one. Tick "Two colonies" to pit nest A against nest B, each with its own trails. The panel reports food collected, ant count and FPS; Restart respawns everything.
Real ants use over twenty distinct trail and alarm pheromones. This model abstracts just two channels — "food found" and "way home" — yet that minimal pair is enough to reproduce the famous double-bridge experiments in which colonies reliably discover the shorter of two routes.
Stigmergy is indirect coordination through the environment rather than direct messaging. An ant changes the world by leaving a pheromone mark, and that mark influences the next ant that passes. No ant has a map or a plan, yet the accumulated marks let the colony behave as if it does, which is exactly what this simulation visualises.
Each step the ant samples the pheromone grid it is currently following — the food trail when searching, the home trail when carrying food — at three points 12 pixels ahead: straight, left and right, separated by the view angle. It rotates toward whichever sample is strongest, then adds a small random nudge so the swarm keeps exploring rather than freezing on one path.
Decay is a multiplier applied to every grid cell each frame, from 0.9 up to 0.999. Values near 0.999 make trails persist for a long time, so even poor routes linger and the colony adapts slowly. Lower values erase trails quickly, forcing constant re-exploration. The interesting behaviour emerges in between, where good paths are reinforced faster than they fade.
It captures the core mechanism faithfully: deposit, follow, and evaporate. Real ants add factors this model omits, such as trail recruitment intensity, terrain, and many chemical types. The double-trail scheme and exponential decay here mirror the standard textbook model and the ant colony optimisation algorithm, so it is a sound conceptual rather than biologically exact representation.
Ant colony optimisation, or ACO, is a class of algorithms inspired by exactly this foraging behaviour. Virtual ants build candidate solutions, deposit pheromone in proportion to solution quality, and evaporation discounts older choices. Run on a graph instead of an open field, the same deposit-follow-decay loop solves routing problems like the travelling salesman, which is why this sim sits in the algorithms family.