Self-organized task specialization and task allocation displayed by a controller evolved from first principles using Grammatical Evolution (cf. S3 Video and evolved rules shown in S2 Table).

<p>(a) Robot densities in the experimental arena of as a function of time (average of 30 runs). Despite having identical controllers, robots segregate quickly between the source and cache areas, thereby avoiding the costly traverse of the slope. (b) Robot trajectory on the arena and cache size in a typical evaluation run. All robots first move to the source to collect items, but after 500–1000 s into the simulation, the robot teams self-organize to have two droppers pushing items off the slope and two robots collecting items from the cache, without these tasks having been explicitly rewarded during the evolutionary runs.