Why Build a Thousand Stupid Robots Instead of One Clever One? Nothing Important Breaks
By the BrainSnail editorial team. How these articles are written and checked, and how to tell us when one is wrong.
Large numbers of simple machines following local rules can accomplish tasks together that none of them could manage alone, with no leader and no central plan. The approach copies insects deliberately and inherits both their strengths and their limits.
What the approach is
The design principle is that each unit is simple, cheap and interchangeable, senses only its immediate surroundings, communicates only with nearby units, and follows rules that refer to nothing beyond what it can perceive. No unit knows the overall plan, no unit directs the others, and the behaviour of the group as a whole emerges from those local interactions rather than being specified anywhere. That arrangement is copied explicitly from social insects, where colonies build, forage and defend without any individual understanding the project, and the research field takes both its methods and a great deal of its terminology from biology.
What the arrangement buys
The advantages follow directly from having no central component:
- •No single failure stops the task, since every unit is replaceable
- •Performance degrades gradually as units are lost rather than stopping
- •The group scales up by adding units without redesigning anything
- •Units can be manufactured cheaply in quantity
- •Communication requirements stay small regardless of group size
- •The group covers an area in parallel rather than sequentially
What it is actually used for
Practical deployment lags the research considerably and several applications are established. Warehouse robots move shelves to picking stations in large numbers, coordinating through a central system rather than purely locally but sharing the interchangeability. Agricultural work with many small machines instead of one enormous one is being trialled and reduces soil compaction substantially. Environmental monitoring uses floating and underwater units covering an area together. Search after disasters is a standing proposal, since many cheap units entering a collapsed structure risk nothing and cover more ground. Coordinated aerial displays using hundreds of lit aircraft are the most visible commercial use. Construction by many small units remains largely a laboratory demonstration.
What the insects actually do
The biological systems being copied are worth describing accurately, since the engineering borrows specific mechanisms rather than a general idea. Ants find short routes by laying a trail chemical that evaporates, so a shorter route is traversed more often, is reinforced more frequently and outcompetes a longer one without any ant comparing them. Termites build by depositing material where material has already been deposited, which produces pillars and then arches with no plan anywhere. Honeybees allocate foragers between sources by dancing with a vigour reflecting quality, which is a market without prices. Each of these is a specific rule producing a specific global outcome, and algorithms named after each are used well beyond robotics.
Why it is harder than it looks
Designing local rules that produce a wanted global behaviour is the central difficulty and there is no general method for doing it. Rules are usually found by trial, by simulation or by copying a biological system, and proving that a set of rules will always produce the intended result is frequently impossible. Groups can settle into stable wrong states, gridlock, or oscillate. Behaviour that works in simulation frequently fails with real hardware, where sensing is noisy, communication drops and units differ slightly. Debugging is difficult, since a fault is a property of the interaction rather than of any unit. And demonstrating reliability sufficiently to deploy such a system anywhere safety matters remains largely unsolved.
The takeaway
Each unit is simple, senses only its surroundings, communicates only with neighbours and follows rules referring to nothing global, so the group behaviour emerges rather than being specified. That removes any single point of failure, degrades gradually and scales by adding units. Warehouses, agriculture, monitoring and aerial displays use it. Designing local rules that reliably produce a wanted global result has no general method.