Simple individual behavioural rules for improving the collective behaviours of robot swarms

Swarm robotics is an ongoing area of research that is expected to revolutionise various real-world domains such as agriculture and space exploration. Swarm robotics systems are composed of a large number of simple and autonomous robots. Each robot locally interacts with other robots and with the environment following a set of behavioural rules. These individual interactions enable the swarm to exhibit interesting collective behaviours and to accomplish specific tasks. The main challenge in designing robot swarms is to determine the behavioural rules that each robot should follow so that the swarm as a whole can perform the desired task. The performance of robot swarms in a given task depends on the designer's choice of appropriate individual behavioural rules. In this thesis, we investigate simple individual behavioural rules for improving the performance of robot swarms in two major tasks. Using simple behavioural rules makes the designed solutions possibly usable with simpler platforms such as micro- and nanorobots.

The first task we address is known as the best-of-n decision problem where the swarm is required to select the best option among n available alternatives. Solving the best-of-n decision problem is considered to be a fundamental cognitive skill for robot swarms as it influences the swarm's success in other tasks. In this thesis, we introduce individual behavioural rules to improve the performance of robot swarms in the best-of-n problem. Through these rules, robots vary their interaction strength over time in a decentralised fashion to balance the acquisition and the dissemination of information. The proposed behavioural rules allow swarms of simple noisy robots with constrained communication to limit the effect of individual errors and make highly accurate collective decisions in a predictable time.

In some scenarios where the best option changes over time, the swarm is required to switch its decision accordingly. In this thesis, we introduce individual behavioural rules through which the robots process new information and discard outdated beliefs. These behavioural rules enable robot swarms to adapt their decisions to various environmental changes, including the appearance of better choices or the disappearance of the current swarm's choice. Our analysis shows that relying on local communication is more favourable for achieving adaptation. This result highlights the benefit of the local sensing and communication characterising biological and artificial swarms.

The second task we address in this thesis is the collective resource collection task. In this task, the robots are asked to retrieve objects that are clustered at unknown locations in the environment. We address this task because of its numerous potential real-world applications. In many of these applications, the objects to collect are assigned different importance or value. In this thesis, we introduce a bio-inspired individual behaviour that allows robot swarms to perform quality-based resource collection. Similarly to foraging ants, in our proposed behaviour, the robots coordinate their collection efforts by laying and sensing virtual pheromone trails. The use of pheromone trails offers an advantageous implementation of the memory and communication capabilities necessary for the efficient collection of clustered objects. The proposed behaviour allows robot swarms to satisfy various collection objectives and achieve an optimal resource collection behaviour in the case of relatively small swarms.

In this thesis, we analyse swarm robotics systems using both minimalistic tools such as stochastic and multi-agent simulations, and more advanced tools such as physics-based simulations and real robot experiments. Using these tools, we demonstrate the effectiveness of the proposed individual behavioural rules in improving the performance of robot swarms in the addressed tasks. The results we present in this thesis are of potential interest to both engineers designing robot swarms, and biologists investigating the behavioural rules followed by individuals in living collective organisms.