Question 1

What is Ant Colony Optimization and what problem does it solve here?

Accepted Answer

Ant Colony Optimization (ACO) is a metaheuristic inspired by how real ants find short paths between their nest and food using pheromone trails. In this simulation it solves a multi-stop delivery routing problem: a drone must visit every delivery point exactly once and return to base, which is the classic Travelling Salesman Problem (TSP). TSP is NP-hard, so for anything beyond a handful of stops it is impractical to check every possible ordering. ACO instead runs many simulated ants that build candidate routes probabilistically, reinforcing whichever edges tend to appear in short routes, so the population of routes gets better over successive colonies without ever proving optimality.

Question 2

How does the ACO transition-probability rule decide where each ant goes next?

Accepted Answer

At each step, an ant standing at delivery point i chooses its next unvisited stop j with probability proportional to [τ(i,j)]^α × [η(i,j)]^β, where τ(i,j) is the pheromone level on edge (i,j) and η(i,j) = 1/distance(i,j) is the heuristic desirability — nearer points look more attractive. α controls how strongly the ant follows accumulated colony experience (pheromone), while β controls how strongly it follows pure greedy distance. The ant then makes a weighted random (roulette-wheel) draw among all unvisited candidates using these combined scores, so it usually — but not always — picks a promising edge, which keeps the colony exploring.

Question 3

Why does pheromone evaporate, and what does the evaporation rate ρ control?

Accepted Answer

After every colony finishes its tours, all pheromone values are multiplied by (1 − ρ) before new deposits are added. Without evaporation, pheromone would only ever accumulate, and whichever edges got lucky early would dominate forever — trapping the search in a mediocre solution. Evaporation lets weak or stale trails fade so the colony can keep exploring alternative routes. A high ρ forgets history quickly and explores more but converges more slowly and noisily; a low ρ remembers longer and converges faster but risks locking onto an early, suboptimal route (premature convergence).

Question 4

How is pheromone deposited, and why do shorter routes deposit more?

Accepted Answer

After evaporation, every ant in the colony deposits pheromone of magnitude Q / L on each edge of the tour it built, where L is that ant's total route length and Q is a fixed constant. Because the deposit is inversely proportional to length, an ant that found a short route reinforces its edges much more strongly than an ant that found a long, inefficient one. Over many colonies this differential reinforcement is what turns a purely random search into one that concentrates pheromone — and therefore future ant traffic — onto the edges that repeatedly show up in short routes.

Question 5

What do the α and β sliders do at their extremes?

Accepted Answer

Setting α = 0 makes ants ignore pheromone entirely and behave like a greedy nearest-neighbour-style heuristic driven only by η (distance), so no colony memory forms and there is little improvement over time. Setting β = 0 makes ants ignore distance entirely and follow pheromone alone, which can cause the whole colony to reinforce one early, possibly poor, route very quickly (stagnation). The classic balance uses a moderate α (around 1) with a stronger β (around 2–5) so early exploration is distance-aware, while pheromone still lets good structure compound across colonies.

Question 6

Why does the best-route length sometimes plateau instead of always improving?

Accepted Answer

The chart tracks the best route length found so far, which by definition can never get worse — it is a running minimum. It plateaus whenever no ant in the most recent colonies has managed to beat the current champion route, which is expected: as pheromone concentrates on good edges, most ants converge toward similar tours and genuinely novel improvements become rarer. Long plateaus usually mean the colony has settled near a local optimum for the current parameters; increasing the evaporation rate, the ant count, or the exploration exponent β can sometimes shake it loose to find a shorter route.

Question 7

How does this differ from an exact TSP solver?

Accepted Answer

An exact solver (branch-and-bound, dynamic programming, or integer programming) can guarantee the true shortest possible route, but its running time grows explosively with the number of stops — dynamic programming already needs roughly n²·2ⁿ operations, which becomes infeasible well before n = 30. ACO gives up the optimality guarantee in exchange for scalability: it produces good, often near-optimal, routes for much larger instances in a fixed amount of computation, which is exactly the trade-off real delivery and logistics systems make when routing dozens or hundreds of stops.