Further Investigations on the Characteristics of Neural Network Based Opinion Selection Mechanisms for Robotics Swarms

Abstract

Collective decision-making is a process that allows a group of autonomous agents to make a decision in a way that once the decision is made it cannot be attributed to any agent in the group. In the swarm robotics literature, collective decision-making mechanisms have generally been designed using behaviour-based control structures. That is, the individual decision-making mechanisms are integrated into modular control systems, in which each module concerns a specific behavioural response required by the robots to respond to physical and social stimuli. Recently, an alternative solution has been proposed which is based on the use of dynamical neural networks as individual decision-making mechanisms. This alternative solution proved effective in a perceptual discrimination task under various operating conditions and for swarms that differ in size. In this paper, we further investigate the characteristics of this neural model for opinion selection using three different tests. The first test examines the ability of the neural model to underpin consensus among the swarm members in an environment where all available options have the same quality and cost (i.e., a symmetrical environment). The second test evaluates the neural model with respect to a type of environmental variability related to the spatial distribution of the options. The third test examines the extent to which the neural model is tolerant to the failure of individual components. The results of our simulations show that the neural model allows the swarm to reach consensus in a symmetrical environment, and that it makes the swarm relatively resilient to major sensor failure. We also show that the swarm performance drops in accuracy in those cases in which the perceptual cues are patchily distributed.