Responsive leadership: robotic leaders perform better in recruitment tasks if they adapt to the group's stress level

Consistent behavioral differences, or behavioral syndromes, exist in a variety of species. Groups of guppies (Poecilia reticulata) for example can be composed of bold or shy fish, active or lethargic and more or less sociable ones.

For a given task, such as leading the fish to a food source, the optimal strategy to reach the goal might vary significantly depending on the group's personality composition. Biomimetic robots have been shown to be a powerful tool in studying collective behavior of fish.

While the behavior of live fish may depend heavily on the group's composition, the social interactions of a robot are fully controllable. Social cues can be injected into and uncoupled from the dynamics of the group. Over the last years we have developed robotic fish that can be integrated into live fish shoals.

The system tracks all fish in the tank and robots can be controlled in closed-loop with its environment. In the talk I will present our latest research in which we developed robots that can classify fish behavior into behavioral classes, such as "fear" or "attraction".

We asked whether feeding back this information to the robot's behavior could improve its performance in a recruitment tasks. We find that robots that adapt are indeed better leaders and that stress-related behavioral observations make up most of this effect.