Human-Automation Interaction Lab

Research in the Human-Automation Interaction Lab focuses on the role of human attention, memory, decision making, and trust in advanced technological systems. Previous work on this topic examined automation in the context of aviation and air traffic control, but our current focus in on supervisory control of unmanned vehicles and on human-robot interaction. Our goal is to leverage cognitive theory and neuroergonomic methods to support effective and safe human-automation interaction in these systems. In particular, we have examined the benefits of adaptive and adaptable automation as approaches to balance operator cognitive workload, enhance situation awareness, and limit skill degradation.

Adaptive Automation

Automated systems designed without consideration of the human operator role can be associated with unbalanced mental workload, reduced situation awareness, decision biases, mistrust, and complacency (Parasuraman & Wickens, 2008; Parasuraman & Manzey, 2010). We have proposed adaptive automation as a solution to the problems associated with inflexible automation (Parasuraman Sheridan, & Wickens, 2000). In this approach, information or decision support is not fixed at the design stage but presented appropriately depending on context in the operational environment. Context-sensitive adaptive automation is initiated by the system based on critical mission events, operator performance, or physiological state. 

In recent work we have shown the effectiveness of performance-based adaptive automation in supervisory control of multiple unmanned air and ground vehicles (UAV and UGVs). In one study (Parasuraman et al., 2009), participants performed a high workload reconnaissance mission involving four subtasks: (a) UAV target identification; (b) UGV route planning; (c) communications, with embedded verbal situation awareness probes; and (d) change detection. The UAV task was supported by an automated target recognition system that was invoked only if individual operator change detection performance was below a threshold, but not otherwise. Change detection accuracy and situation awareness were higher and workload was lower for adaptive compared to static automation. The results point to the efficacy of adaptive automation for supporting the human operator tasked with supervision of multiple uninhabited vehicles under high workload conditions.

Individual Differences in Human-Automation Interaction: Team Performance and Statistical Modeling

Human operators increasingly work in networked environments involving interaction with automated entities such as decision aids, robots, and unmanned vehicles, as well as with other human team members. Individual operator abilities can be a limiting factor in how effectively multiple agents can be supervised in such environments. Furthermore, such networked systems have complex properties that make prediction of human-system performance difficult, thus necessitating the development of quantitative performance models. Our recent research has addressed both these issues.

In one study (McKendrick et al., in press), we examined whether working memory capacity and communication style were associated with differences in the performance of two-person teams performing a supervisory control task. Two-person teams performed a simulated air defense task with two levels of task load and three levels of reliability of an automated decision aid. Teams communicated and received decision aid messages via chat window text messages. Both verbal and spatial working memory of individual teams members were assessed using span tasks. Working memory capacity was a strong predictor of performance, with average team spatial working memory being the best predictor.  Frequency of team rapport and enemy location communications positively related to team performance, and word count was negatively related to team performance.

In a related study (Ahmed et al., in press), we examined the challenging problem of modeling the interaction between individual differences and decision-making performance in networked human-automation system tasks. Analysis of single-operator performance in the same supervisory control task as in the McKendrick et al. (in press) study showed that working memory capacity was a strong mediator of the effects of task load and automation message reliability. We modeled these performance effects using three statistical approaches:  classical linear regression, nonparametric Gaussian processes (GPs), and probabilistic Bayesian networks (BNs). We found that each of these approaches can help predict performance on networked human-automated systems by linking expected operating conditions and performance from real experimental data to observable cognitive traits like working memory capacity. Such statistical models, especially GP models, also help predict performance for independent variables for which performance data are not available, e.g., to predict performance when operators have to supervise a larger number of unmanned vehicles.