A piston in a cylinder with gas usually moves when the gas pushes on it. But there is a piston that moves when the space in front of it is empty. Every five seconds, it checks the region near its face, and if no one is there, it takes a step inward, compressing the gas without direct mechanical force. This is how an information engine works: it converts knowledge about particle positions into useful work.

In the new experiment, the working substance was not an ordinary gas but a gas of self-propelled robots. These brush-like robots vibrate and crawl across the arena like living creatures, accumulating near the walls. It is this accumulation that changes the rules of the game. For an ordinary gas, the average work per measurement follows a simple formula, and the maximum is achieved when the probability of finding the region empty is about 37%. For an active gas, things are different.

At low robot density, it is advantageous to check large regions: they are more often empty, and the piston takes a large step. But when there are many robots, a boundary layer forms near the wall, and the optimal strategy switches. Now it is better to check a small region right next to the piston, inside this layer, where emptiness occurs more frequently. The transition occurs when the area fraction is about 8%, which is like having eight robots, each the size of a palm, in a room of 30 square meters.