Collective construction usually requires either centralized coordination or complex individual agents. A central planner knows the target shape and assigns each robot a position. Alternatively, each robot carries enough intelligence to reason about the global structure. Both approaches scale badly: the planner becomes a bottleneck, and the individual complexity makes each unit expensive.
Huddle takes the third path: parallel shape assembly using decentralized, minimalistic robots. Each robot follows only local rules — sensing nearby neighbors, comparing local configurations against a shared shape description. No robot knows its global position. No controller directs traffic. The target shape emerges from parallel local decisions about where to attach and where to wait.
The key mechanism is that the shape description is distributed, not centralized. Each robot carries the same compact representation of the target structure. When a robot encounters the boundary of the growing assembly, it checks whether its local neighborhood matches a valid attachment site in the description. If yes, it locks in place. If no, it moves on. The global shape assembles correctly because every attachment decision is locally validated against the same global template.
This is parallel construction without a foreman. The foreman's knowledge is encoded in the template that every worker carries. The efficiency comes from the fact that many robots can simultaneously evaluate different attachment sites — the assembly front grows in parallel rather than sequentially. The cost is that certain shapes are harder to describe locally than globally. The representation has to make global structure legible from local views.