Automated Calibration of a Biomimetic Space-Dependent Model for Zebrafish and Robot Collective Behaviour in a Structured Environment

Abstract

Bio-hybrid systems made of robots and animals can be useful tools both for biology and robotics. To socially integrate robots into animal groups the robots should behave in a biomimetic manner with close loop interactions between robots and animals. Behavioural zebrafish experiments show that their individual behaviours depend on social interactions producing collective behaviour and depend on their position in the environment. Based on those observations we build a multilevel model to describe the zebrafish collective behaviours in a structured environment. Here, we present this new model segmented in spatial zones that each corresponds to different behavioural patterns. We automatically fit the model parameters for each zone to experimental data using a multi-objective evolutionary algorithm. We then evaluate how the resulting calibrated model compares to the experimental data. The model is used to drive the behaviour of a robot that has to integrate socially in a group of zebrafish. We show experimentally that a biomimetic multilevel and context-dependent model allows good social integration of fish and robots in a structured environment.