Building a safe robot for behavioral biology experiments

A robot accepted by animals as conspecific is a very powerful tool in behavioral biology, particularly in studies of gregarious animals. However in experiments where animals and robots share the same physical environment, there is always a risk of a robot accident that can lead to animal injuries. Safety regulations have to be developed to guide the design of an intrinsically safe robotic hardware and software. Currently this question is not addressed in experimental biology. In this paper, we present our efforts to build a safe robot for experimentation with domestic chickens. The methodology we use is based on robot safety studies done in the field of physical humanrobot interaction. The safety elements were introduced in the mechanical design of the robot, its control system and into the experimental environment. We show how in particular cases, when local information available to the robot is insufficient to detect an abnormal situation, the global information provided by the external vision system can be used by a novelty detection system based on extreme value theory. We believe that this study can be useful for robotic researchers providing robots for biological studies, as the concepts presented are universal and can be applied to other types of animals.