User evaluation of a GUI for controlling autonomous persistent surveillance team

Abstract

In future military missions, there will be many sensor assets collecting much important information about the environment. User control over surveillance assets is important to ensure that the specific data collected is appropriate for the current mission. Unfortunately, previous work has shown that individual users cannot effectively control more than about four assets, even if the assets have significant autonomy. In the ACCAST project, we hypothesized that by including autonomous teamwork between the assets and allowing users to interact by describing what the team as a whole and specific sub-teams should do, we could dramatically scale up the number of assets an individual user could effectively control. In this paper, we present the results of an experiment where users controlled up to 30 autonomous assets performing a complex mission. The assets autonomously worked together using sophisticated teamwork and the user could tell sub-teams to execute team oriented plans which described the steps required to achieve a team objective without describing exactly which asset performed which role and without having to specify how the team should handle routine information sharing, communications and failure circumstances. The users, soldiers from Fort Benning, were surprisingly good at managing the assets and were all able to complete the complex mission with extremely low friendly and civilian casualties.