Meeting Abstract
Previous research on swarm behaviour has shown that local interactions among neighbours can lead to stable movement patterns at the group level without a centralized control strategy. However, reverse-engineering the structure and dynamics of these interactions (either to understand the collective motion of animals or to achieve a desired collective behaviour in multi-robot systems) has been challenging. To begin to elucidate the relationship between local and global, we ask what is the minimum number of neighbours each member needs to take into account for a group to achieve a high performance? To address this question, we perform computer simulations in a three-dimensional environment with obstacles. Our agent interaction model expands on previous studies where each group member is programmed to choose between three simple behaviours: avoidance, alignment and attraction. A novel winner-takes-all approach is implemented for decision making with highest priority assigned to avoidance. The decision whether to align or attract is determined depending on the number of neighbours found in the near field, the probability of aligning increases with more neighbours. The performance of the group is evaluated using two metrics: cohesion calculated as the number of splits occurred within the group (fewer splits indicate higher cohesion) and mobility calculated as the coverage rate of the environment while avoiding obstacles (higher coverage indicates higher mobility). Our preliminary results suggest that independent of group size, agents can stay together and move around effectively by keeping only 6 neighbours in their near field. Next, we plan to conduct biological experiments to test whether a similar sensory feedback mechanism exists in fish schools and to develop new bio-inspired control strategies for multi-robot systems.
