A Comparative Study on Decision Making Mechanisms in a Simulated Swarm of Robots

Enhancing Robustness of Swarm Robotics Systems in a Perceptual Discrimination Task

“…The results achieved by this study show that, regardless of the difference in quality, groups designed to perform optimally in the Random type of environments (see Figure 1, Random) experience a performance drop when they are post-evaluated in the Off-diagonal and Stripe environment (see Figure 1, Off-diagonal, and Stripe). This observation has been corroborated by other recent studies [12], [16], [17] which report the same type of performance drop in spite of the fact that they employ artificial neural network as robots controllers to improve the robustness of the collective response. The main focus of this study is to overcome the limitations illustrated in [15], [12], [16], [17] by developing individual decision-making mechanisms underpinning a collective response that allow a swarm of robots to perform sufficiently well in all the nine types of floor distribution patterns illustrated in Figure 1.…”

“…This observation has been corroborated by other recent studies [12], [16], [17] which report the same type of performance drop in spite of the fact that they employ artificial neural network as robots controllers to improve the robustness of the collective response. The main focus of this study is to overcome the limitations illustrated in [15], [12], [16], [17] by developing individual decision-making mechanisms underpinning a collective response that allow a swarm of robots to perform sufficiently well in all the nine types of floor distribution patterns illustrated in Figure 1. In order to achieve this objective, we focus on multiple elements such as the type of individual random walk used by the robots to explore the arena, the structure of the neuro-controller, as well as on the characteristics of the communication strategy used to exchange individuals opinions.…”

“…A notable feature of the ER approach is the automation of the design process, which significantly reduces the influence of designer assumptions. Recent research work [14] provides evidence that the ER approach outperformed the handcoded approach with respect to robustness, scalability, and adaptability of the collective response of the group.…”

“…We found out that, this latter element is the one that allowed us to achieve an important improvement in terms of robustness of the collective decision with observing specific floor patterns. Particularly, we found out that only when the communication events happens between a robot receiver and a randomly chosen (rather than the closest as in [15], [12], [16], [17]) emitter among those within communication range, no performance drop is observed while moving from Random to all the other nine floor patterns. We show that the superior robustness observed in group in which the communication happens between a robot receiver and a randomly chosen emitter can be attributed to a more effective circulation of opinions among both the spatially close and distant robots, thereby maintaining a high accuracy rate in the decision process throughout the eight floor patterns not experienced by the robots during the design phase.…”

Review of Collective Decision Making in Swarm Robotics

“…[34], [36], [39], [40], [41] Automatic Approach / Evolutionary Robotics A neural network synthesised using evolutionary optimisation technique, used as an individual controller.…”