Improving shepherding tasks through destination-orientated actions

The shepherding task is a classical example of swarm intelligence which often utilizes one intelligent agent (sheepdog) to herd a flock of simple agents (sheep) towards a predefined destination. During the shepherding, the sheepdog usually employs two actions with one for collecting all sheep and another one for driving the flock to the destination, and switches between the two constantly according to the aggregation degree of the flock. Conventional collecting action, however, simply compels the sheepdog to run towards a sheep that is farthest away from the flock's center, while taking no account of the sheep's location relative to the destination. This paper proposes an alternative collecting action which selects a sheep from the flock as target that has maximum sum of distances from the sheep to both the flock's center and the destination. Moreover, in order to speed up the switch between the collecting and driving actions, the aggregation level of the sheep flock will be measured in terms of an elastic circular sector centered at the destination, rather than a rigid circular surrounding the sheep flock. Numerical experiments demonstrate that the destination-orientated actions can facilitate a more efficient gathering of scattered sheep, and hence, improve the efficiency of the shepherding tasks.