Topologica linteractions in a multi-layered flocking system

Abstract: With the multi-layered flocking system it is possible to simulate flocks that contain different types of agents that can be of various different sizes (variations in bounding radius and height). In the original implementation, the multi-layered flocking system uses a metric distance to find the nearest-neighbours of agents. However, results from real life field studies suggest that animals interact with each other in a flock using a topological distance. The goal of this thesis is therefore to implement a version of the multi-layered flocking system that uses a topological distance for interaction between agents. This is done by adapting two methods that are used to find the k-nearest neighbours (kNN), namely the original spatial partitioning method (OSP method) and the enhanced spatial partitioning method (ESP method), to work with the multi-layered flocking system. The aim is to compare the performance of these methods in terms of query time for four different flocking scenarios (standard, obstacle, follow and steer away). The implementation contains two types of agents of two different sizes. In the standard scenario all agents move together as a flock. The obstacle scenario is similar to the standard scenario with the addition that the simulation space contains stationary obstacles. In the follow scenario the smaller sized agents follow the bigger sized agents, and in the steer away scenario the smaller sized agents steer away from the bigger sized agents. An evaluation of how different numbers of kNN affect the collective motion (polarization, extension and frequency of collisions) of the flock in the four different scenarios is also done. The evaluation was performed by implementing the multi-layered flocking system in the Unity game engine, and running simulations with flocks of different sizes (125-3125 agents) and using different numbers of interacting kNN (k=5,10,15,20) for each of the scenarios. The results show that the ESP method on average is at least twice as fast compared to the OSP method in all four flocking scenarios, and the improvement in performance in query time did not differ much between the scenarios. Moreover, a value of k=10 was shown to be a good compromise between having fast kNN query times for the ESP method, but still having flocks of agents moving in a collective manner.