An Autonomous Robotic Platform to Manipulate Collective Behavior in Ants

Eusocial life is characterized by division of labour, collective decision making and self organization, and regarded as the highest form of social organisation in groups. Ants are a model organism for research in collective behavior and the evolution of eusociality. Recently, mobile robots have been developed as an experimental tool to investigate animal behavior in manipulative interactive experiments with mammals, fish, birds and bees. Manipulating the collective behavior of an eusocial organism at the individual level is instrumental to study the mechanisms of self-organization. In this thesis, the Antbot, a robotic manipulation platform, is developed with an ant-sized mobile dummy to deliver a tactile stimulus to a target individual without interacting with any other ant. The system is integrated with a state-of-the-art tracking system to provide real-time visual feedback for teleoperated and autonomous manipulation.

We use the Antbot in an experiment to investigate the social factors that influence the levels of individual aggression and responsiveness. Aggression is a central behavior in animals, particularly in eusocial insects, as it serves to maintain the colony integrity by protecting resources such as brood, food, territory and workforce. We found that aggression and responsiveness to tactile stimuli depend on the previous activity of the ants, the social context, and the individual task profile. We also demonstrated a habituation effect in individuals that were previously more active. As a milestone towards automated experimentation, a navigation algorithm for socially aware navigation in dynamic crowded environments is developed and implemented on the platform. A software framework is developed for real-time processing of social parameters, such as interaction network measures, that provide the basis for experimental automation with easily re-programmable behavior for the robot. The autonomous operation of the Antbot platform is tested with a proof-of-concept experiment.