Evolutionary Optimisation of Neural Network Models for Fish Collective Behaviours in Mixed Groups of Robots and Zebrafish

Abstract

Animal and robot social interactions are interesting both for ethological studies and robotics. On the one hand, the robots can be tools and models to analyse animal collective behaviours, on the other hand the robots and their artificial intelligence are directly confronted and compared to the natural animal collective intelligence. The first step is to design robots and their behavioural controllers that are capable of socially interact with animals. Designing such behavioural bio-mimetic controllers remains an important challenge as they have to reproduce the animal behaviours and have to be calibrated on experimental data. Most animal collective behavioural models are designed by modellers based on experimental data. This process is long and costly because it is difficult to identify the relevant behavioural features that are then used as a priori knowledge in model building. Here, we want to model the fish individual and collective behaviours in order to develop robot controllers. We explore the use of optimised black-box models based on artificial neural networks (ANN) to model fish behaviours. While the ANN may not be biomimetic but rather bio-inspired, they can be used to link perception to motor responses. These models are designed to be implementable as robot controllers to form mixed-groups of fish and robots, using few a priori knowledge of the fish behaviours. We present a methodology with multilayer perceptron or echo state networks that are optimised through evolutionary algorithms to model accurately the fish individual and collective behaviours in a bounded rectangular arena. We assess the biomimetism of the generated models and compare them to the fish experimental behaviours.

Modelling zebrafish collective behaviours with multilayer perceptrons optimised by evolutionary algorithms

“…Namely, small circular loops can appear in all simulated cases (especially here with NSGA-III cases) when simulated agents performs U-turns to catch up with the rest of the group. Previous analyses in 36 showed that this effect cannot be easily mitigated even when the fitness function takes into account agents angular speeds.…”

“…makes them suitable to model control problems 39 . We consider MLP with only one hidden layer of 100 neurons (using a hyperbolic tangent function as activation function) as in 36 .…”

“…Our approach models fish trajectories resulting from social interactions in a homogeneous environment bounded by walls. Very few models of fish collective behaviours take into account the presence of walls 22,23,30,36 .…”

“…We compute a biomimetism score using the similarity measure defined in 30,36,42 to quantify the realism of the behaviour of a group of simulated agents. It compares the behaviours of this group with the behaviours of the experimental fish averaged across all 10 experimental trials (Control case e c ).…”

“…We calibrate the MLP models describing agent behaviour to match as close as possible the dynamics of fish in groups of 5 individuals during 30 minutes simulations (15 time-steps per seconds, i.e. 27000 steps per simulation) using a methodology inspired from 36,42 . This is achieved by optimising the connection weights of the MLP using evolutionary algorithms that iteratively perform global optimisation (inspired by biological evolution) on a defined fitness function so as to find its optimal value 45,46 .…”