Research On Bio-inspired Shepherding Task Based On Improved Aggregation Strategies

Swarm intelligence studies the intelligent behavior of a large number of simple individuals spontaneously emerging through interaction.Shepherding tasks is a classical swarm intelligence model.Its purpose is to imitate the behavior of sheepdog driving sheep in nature,and to study how a small number of intelligent individuals effectively control a large number of simple individual actions.The action strategy of sheepdog can be widely used in the control of robots or UAVs,crowd evacuation,earthquake relief and maritime rescue.This thesis focuses on self-propelled particle model,which is one of the most basic models of shepherding task.The goal of this model is to use a single sheepdog to drive a group of sheep randomly distributed in the space to a designated area(destination).In the process of driving sheep,by observing the aggregation degree of sheep,the sheepdog alternately uses two kinds of action strategies,namely collecting and driving strategies.Among them,the collecting strategy gathers the scattered sheep into a close group,while the driving strategy drives the gathered sheep towards the destination.However,the traditional collecting strategy chooses the sheep farthest from the sheep center as the action target each time,ignoring the relative position of the target sheep,the sheep center and the destination,so it is easy to disperse the sheep or make the sheep deviate from the target area in the process of collecting sheep,thus increasing the total time cost.In order to improve the efficiency of shepherding task,this thesis proposes two improved aggregation strategies,namely the maximum angle strategy and the maximum perimeter strategy.The maximum angle aggregation strategy can solve the problem of scattering sheep in the process of aggregation.The principle of this strategy is to calculate the angle between the current position of the shepherd dog and the center of the sheep and the line of each sheep,and select the sheep with the largest angle as the action target.This strategy can protect the gathered sheep from being scattered by the sheepdog.The experimental results show that the strategy is superior to the traditional strategy in time steps,trajectory distance and dispersion degree,and the most important time step index is increased by 12.70% on average.The maximum perimeter aggregation strategy can solve the problem of gathering sheep too slowly.The principle of this strategy is that every time the sheepdog gathers sheep,it always chooses the sheep with the largest distance from the center of the sheep and the destination as the action target,avoiding the sheep close to the destination.At the same time,in order to improve the switching frequency between the two strategies,this thesis sets the fan-shaped area with variable angle and destination as the evaluation standard of sheep aggregation degree.This improvement can effectively avoid the sheepdog to drive the sheep near the target area,and save the time cost of sheep task.Experiments show that the maximum perimeter aggregation strategy is superior to the traditional strategy in time steps,trajectory distance and dispersion degree,especially the time steps index is 40.07% higher than the traditional model.Finally,in order to further verify the effectiveness of the two improved strategies,this paper conducted experiments on large-scale sheep shepherding task,and discussed the limitation of single sheepdog control ability after the number of sheep increased.