Analyzing the effects of dynamic task allocation on human-automation system performance

Thesis: Ph. D., Massachusetts Institute of Technology, Department of Aeronautics and Astronautics, 2015.Cataloged from PDF version of thesis.Includes bibliographical references (pages 123-134).Modem complex aerospace systems employ flight deck automation to increase the efficiency and safety of systems while reducing operator workload. However, too much automation can lead to overtrust, complacency, and a decrease in operator situation awareness. In an attempt to prevent these from occurring, the operator and the automation often share responsibility for performing tasks. The tasks allocated to each agent are rarely fixed; instead, they can be dynamically re-allocated throughout operations based on the state of the operators, system, and environment. This thesis investigates how dynamic task re-allocation has been implemented in operational aerospace systems, and investigates the effect of control mode transitions on operator flying performance, visual attention, mental workload, and situation awareness through experimentation and simulation. This thesis reviews the dynamic task allocation literature and discusses the ways in which the concept can be implemented. It highlights adaptive automation, in which the dynamic re-allocation of tasks is initiated by the automation in a manner that is adaptive - in response to the state of the operator, system, and environment - and workload-balancing - with the purpose of keeping the operator in control as much as possible while remaining at a moderate level of mental workload. Adaptive automation is enthusiastically supported in the literature; however, for reasons discussed, it has not been deployed in any operational civilian aerospace system. In the experiment, twelve subjects sat at a fixed-base lunar landing simulator and initiated transitions between automatic and two manual control modes. Visual fixations were recorded with an eye tracker, and subjects' mental workload and situation awareness were measured using the responses to a secondary two-choice response task and a tertiary task of verbal call-outs of the vehicle state, respectively. Subjects were found to re-allocate attention according to the priority of tasks: during mode transitions from autopilot to two-axis manual control the percent of total attention on the attitude indicator (which was required for the primary flying task) increased 14% while attention on instruments required for the secondary and tertiary tasks decreased 5%. Subjects' conception of task priority appeared to be influenced by instructions given during training and top-down and bottom-up properties of the tasks and instrument displays. The attention allocation was also affected by the frequency of control inputs required. The percent of attention on the attitude indicator decreased up to 13% across mode transitions where the flying task was not re-allocated because the pitch guidance rate-of-change decreased from -9 to 0 °/s throughout the trial. Consequently, fewer control inputs and less attention were necessary later in the trial. An integrated human-vehicle model was developed to simulate how operators allocate attention in the lunar landing task and the effect this has on flying performance, mental workload, and situation awareness. The human performance model describes how operators make estimates of the system states, correct these estimates by attending and perceiving information from the displays, and use these estimates to control the vehicle. A new attention parameter - the uncertainty in operators' estimates of system states between visual fixations - was developed that directly relates attention and situation awareness. The model's attention block was validated against experimental data, demonstrating an average difference in the percent of attention </-3.6% for all instruments. The model's predictions of flying performance, mental workload, and situation awareness were also qualitatively compared to experimental data.by Aaron William Johnson.Ph. D