Solution Space-Based Complexity Metric for ATC

Air Traffic Controller (ATCo) workload is considered to be one of the main limitations in the capacity of the current Air Traffic Management system. The limited options available for implementing changes in this system and their possible effects on ATCo workload have made it subject of many investigations. (Hilburn 2004) One of these topics is the analysis of ATCo task demand load (the objective, subject-independent aspects of workload), which finds its motivation in the reduced costs of implementation, its non obtrusive nature, and its possible implementation as an evaluation metric for ATC sector design. Several sector complexity metrics have been proposed for estimating ATCo task demand load, being Aircraft Density and Dynamic Density by far the most studied ones. Reliable prediction of ATCo workload based on objective metrics has, however, showed to be a great challenge. Hilburn (2004) elaborated on how complexity metrics developed so far depend heavily on the sector in which they’ve been developed, making its direct use in other sectors unreliable for task load estimation. Hilburn (2004) concludes by saying that no complexity indicator is context-free. This leads us, in general, to the conclusion that a complexity indicator should be context-dependent. Another complexity metric proposed in the literature is one based on the Solution Space Diagram (SSD). Having its foundations in the Velocity Obstacle theory (Mercado et al. 2014), the SSD is a two-dimensional representation that covers all heading/speed combinations possible for a specific aircraft, indicating which velocity vectors offer “safe solutions” and which velocity vectors lead to an impending conflict with another aircraft (van Paassen et al. 2010). Hermes et al. (2009) and van Paassen et al. (2010) have shown that the area that represents unsafe solutions in the SSD have a significant correlation to subjective workload ratings. Hence, apparently, the solution space effectively “captures” elements of the environmental context that predict task difficulty. In a previous study, Abdul Rahman (2014) included the SSD in a comparison of sector complexity metrics that evaluated them in terms of their transferability in capturing dynamic complexity across different controllers and sectors. The comparison was based on results from a human-in-the-loop simulation with two different sectors with varying route layouts. Experiment results revealed that the SSD area metric had a higher correlation with the controller workload ratings that the number of aircraft or the unweighted NASA Dynamic Density metric. It also showed that the weighted NASA metric had higher correlations to workload than the SSD, indicating that the SSD area metric is a better alternative when tuning for a specific sector layout is not feasible. The SSD area metric as used in all previous studies was averaged on all aircraft inside the sector at any given time. This implicitly assumed that each aircraft had an equal influence on workload. This is of course a caveat in all previous studies, since aircraft that just entered the sector cannot pose the same level of difficulty as aircraft that have a conflict that must be solved in a very short period of time. In an attempt to set out a step further towards the elaboration of a context dependent task load metric, this paper aims to analyse the task context of the rerouting task with the Cognitive Task Analysis (CTA) framework. The CTA is a set of methods for identifying cognitive skills, or mental demands, needed to perform a task proficiently (Militello and Hutton 1998). In a study performed by Kilgore et al. (2009), a CTA corresponding to the same controlling goal that falls within the scope of this paper (the rerouting task) was elaborated, and as a result a set of subtasks required to achieve the main controlling goal were identified. The current study proposes a set of metrics that relate to the number of times this set of tasks have to be executed at any given time. The convenience of this approach is the possibility to study the contribution of every subtask to the overall task load by means of correlational analyses. The study hypothesizes that these metrics, in combination with the SSD area metric (to estimate for the difficulty of the task) evaluated only for aircraft pairs that require conflict resolution, provide a better estimate of task load, and therefore a higher correlation to subjective workload ratings. In order to test the hypothesis, human-in-the-loop simulations with subjects who received certain degree of training in ATC were conducted. In order to test the capability of the metrics proposed in this paper to capture task load in different task contexts, two ATC sectors with varying geometries and route layouts were simulated. Results showed that, for different traffic samples, the metrics proposed showed similar or better correlations to subjective ratings of workload than the Dynamic Density metric and SSD area metrics based on the entire aircraft population within the ATC sector. Furthermore, the approach proposed in this paper led to identify a set of subtasks that might have required the most effort from the controller.