Simulating turn-taking behaviours with coupled dynamical recognizers

Abstract

A coupled dynamical recognizer is proposed as a model for simulating turn-taking behaviour. An agent is modelled as a mobile robot with two wheels. A recurrent neural network is used to produce the motor outputs. By controlling this, agents compete to take turns on a two dimensional arena.There are two novel aspects to the present study. First, a dynamical recognizer is not only used for producing motor outputs but also to predict the other agent's behaviour. Second, unlike a mere chasing game, turn-taking behaviour is established only when each agent automatically switches from chaser to evader and vice versa.By using the genetic algorithm technique, we show that turn-taking behaviour is developed between two agents. It is worth noting that turn-taking is established only when an agent fails to predict the other agent's behaviour. In other words, the simultaneous generation of stable (predictable) and unstable (unpredictable) dynamics is inevitable to lead to successive turn-taking behaviour. A relationship between joint attention and prediction will be discussed from this and other related works.