TAKING A NEW STEP IN ADAPTIVE AUTOMATION RESEARCH: APPLIED STUDY OF MODES OF AUTOMATION IN THE COMMERCIAL AIRCRAFT
David B. Kaber
Department of Industrial Engineering, North Carolina State University, Raleigh, NC 27695
ABSTRACT
Low-fidelity simulations of real-world systems have been used extensively in laboratory settings to assess performance, situation awareness (SA) and workload effects of adaptive automation (AA). Different methods to triggering AA or dynamic (system) function allocations (DFA) have been explored and various criteria have been subjectively established to dictate when control allocations should occur. Up to this point in time, the AA research community has invested a great deal of effort in mapping a two-dimensional problem space characterized by the degree of automation of complex systems and when to deliver automated assistance to human operators. In specific, we’ve studied binary approaches to AA involving an artificial form of automation scheduled adjacent tso manual control allocations occurring at regular intervals for pre-defined durations of time across a pre-determined task period. Results of this research have been promising as they demonstrate AA to be useful for enhancing monitoring task performance and for augmenting SA in dynamic control. Unfortunately, the simulations to which AA has been applied have primarily involved psychomotor tasks with few requiring higher-level cognition, such as planning. A science has been made out of relating specific schedules of DFAs to SA in lab studies of, for example, aircraft piloting tasks. This science has been aimed at the important task of establishing a general theory of AA and developing a framework of empirical research results to support it. However, such a theory and the results of laboratory studies are of little value if actual performance, SA, and workload effects of AA do not exist in the context of real-world systems.
It is necessary for AA research to focus on studying the human-machine system performance effects of DFA in high-fidelity simulations of complex systems and to establish methods for applying AA to actual real-world tasks. Furthermore, new studies using high-fidelity laboratory surrogates of real-world systems need to focus on cognitive task performance effects of AA. Preliminary work has been done in this area, but, like historical efforts, it has taken a binary approach to automation. Automation can be considered as a continuum from manual control to full automation with different points along the continuum being characterized by the functions a human and computer controller maintain. Future AA research must approach the concept from this perspective and investigate interactions of various forms of automation, such as batch processing and supervisory control, as part of AA strategies. Although it seems logical to study AA as a continuous concept, this is not a simple task and examining it discretely has been an important beginning involving simple experiments that have generated quick results.
As a first step to address the above needs, an experimental procedure was formulated involving reverse engineering of autoflight systems of an advanced commercial aircraft using a rapid simulation prototyping tool and categorizing actual modes of aircraft automation in terms of a theoretical taxonomy of levels (or degrees) of automation. In addition, fictitious features of autoflight systems were prototyped to enable study of modes of automation not currently available in real aircraft, but that are feasible from a design perspective. The procedure was largely comprised of task and system analyses including an in-depth study of piloting tasks, autoflight system function and trigger design, and the interaction between cockpit controls and displays. It was also necessary to describe the control of the aircraft under various autoflight modes, including identifying what functions the pilot and flight management system are responsible for, and to relate this description to a classical four-stage model of information processing in monitoring, planning, decision-making and action. That is, specific aircraft flight functions were categorized according to these stages and the level of system autonomy was identified on the basis of whether the pilot or autoflight system maintained ownership of a function.
This work is to serve as the basis for empirical studies of the performance and SA effects of using different modes of autoflight in a high-fidelity simulation of an advanced commercial aircraft and the effects of dynamic switching among different modes. It is expected to provide insights into the effectiveness of high degrees of cockpit automation, and the utility of AA in the advanced commercial aircraft for maintaining pilot SA and moderating crew workload.