Does Team Interaction Exploration Support Resilience in Human Autonomy Teaming?

Abstract

Project overview As a team explores interactions, they may find opportunities to expand and refine teamwork over time. This can have consequences for team effectiveness in normal and unexpected situations (Woods, 2018). Understanding the role of exploratory team interactions may be relevant for human-autonomy team (HAT) resilience in the face of synthetic agent rigidity and lack of anticipation (Demir et al, 2019). Team interaction exploration was defined as team interactions with qualities (e.g. content, communication medium) unique to a team’s interaction history (Cooke et al., 2013; Hills et al., 2015). This study examines the relationship between team interaction exploration and HAT performance in multiple remotely-piloted aerial system (RPAS) reconnaissance missions with degraded conditions. The goal of the task was to take good photos of target waypoints. In this task, three teammates are assigned to specific roles: the navigator plans the route using a digital map, the pilot (synthetic) controls the RPAS and selects target waypoints, and the photographer calibrates camera settings to take a good photo of a target waypoint. The synthetic agent was capable of routine team coordination without explicit team player qualities. Teams communicated via a text-chat interface. Seven unique degraded conditions were injected throughout ten missions. Three automation failures disrupted RPAS status information on the photographer’s or pilot’s display, and three autonomy failures disrupted the synthetic agent’s comprehension of waypoint information or caused the agent to move on to the next target before a photo was taken. Finally, a malicious cyber-attack caused the synthetic agent to fly the RPAS to an enemy occupied waypoint. Method Forty-four participants were recruited from a large southwestern university in pairs and formed teams (22 teams) to participate in this study. These participants were either undergraduate or graduate students. This experiment consisted of ten 40-minute missions in total that were carried out over two sessions separated by one-to two-week intervals. After a baseline mission, an automation and autonomy failure was injected into each mission while the team processed target waypoints. The malicious cyber-attack occurred during the final 20-minutes of the tenth mission. This study collected a several measures including measures of team process, physiological measures, and surveys of teamwork knowledge, trust, workload, and anthropomorphism which are not considered in this study. Exploratory team interaction was operationalized as any text-message unique in content, sender, or recipient that was unrelated to routine coordination of target waypoints. Teams were grouped using k-means clustering by their target processing efficiency, number of overcome roadblocks, and mission performance. The three clusters ( K = 3) were comparatively described as low- ( N = 7), middle- ( N = 7), and high-performing ( N = 5) teams. A mixed-factor ANOVA compared the frequency of each team’s exploratory interactions by mission and cluster. Results and discussion High-performing teams were distinguished from middle-and low-performing teams in their ability to maintain high levels of overall performance while efficiently processing targets and overcoming many roadblocks. Middle-performing teams were efficient in overcoming roadblocks but had worse mission performance. The findings indicate that 1) high-performing teams explored team interactions more than middle-performing teams, 2) there was no significant difference in exploration frequency between high-and low-performing teams, and 3) teams explored more in the first session than the second session, with the exception of the final mission. Overall, exploratory team interaction differentiated HAT performance in normal and degraded conditions and should be further examined at other levels of interaction, such as content meaning and interaction patterns.