Improving task allocation in risk-aware robotic sensor networks via auction protocol selection

Abstract

Optimized task allocation in robotic sensor networks is a crucial yet complex problem. A successful allocation must efficiently and rapidly coordinate agents to resolve a greater objective while being resilient to incomplete contextual information. Prior work introduced a two-step task allocation process which first determines a set of suitable robots via an auction, then optimizes the task allocation for these agents with the use of multiobjective evolutionary algorithms. The first step has had success but often yields suboptimal sets in certain conditions, this limitation can be traced back to the auctioning protocol's simplicity, which then constraints the set of potential coalitions sought by the optimizer. We introduce two new auction protocols which are able to choose optimal agents. These protocols are resilient to communication failures, can rapidly determine optimal sets and can select agents based on multiple factors. A valuation system which renders agents aware of the value of entities is also introduced. Finally, a rigorous experimentation is designed to evaluated the performance of the different protocols in multiple scenarios. In the experiment, both proposed protocols in conjunction with the valuation system proved to be efficient methods to select appropriate agents. In both sparse and dense network scenarios, the proposed protocols performed better than the current implementation in terms of selecting appropriate coalitions, at the cost of time and network utilization.