Mechanism Selection for Multi-Robot Task Allocation

The work presented here investigates how environmental features can be used to help select a task allocation mechanism from a portfolio in a multi-robot exploration scenario. In particular, we look at clusters of task locations and the positions of team members in relation to cluster centres. In a data-driven approach, we conduct experiments that use two different task allocation mechanisms on the same set of scenarios, providing comparative performance data. We then train a classifier on this data, giving us a method for choosing the best mechanism for a given scenario. We show that selecting a mechanism via this method, compared to using a single state-of-the-art mechanism only, can improve team performance in certain environments, according to our metrics.