Unmanned aerial vehicles (UAVs) provide many benefits over manned systems, including having smaller platforms, being more cost efficient, enabling longer mission times, allowing flight control that may not be possible with manned aircraft (e.g., higher g forces and deeper enemy penetration), and avoiding loss of human life. UAVs allow for a variety of civilian and military missions not previously available with manned aircraft (Gawron, 1998) and allow these missions to be flown without endangering the pilots, who remain at remote control stations (Mouloua, Gilson, & Hancock, 2005).
Currently the U.S. Army flies two of its UAVs, the Hunter and Shadow (short- to medium-range tactical reconnaissance UAVs), with two operators for each vehicle. In order to increase the number of UAVs flying without increasing personnel requirements, there is interest in having a single pilot fly one UAV or even two UAVs concurrently. Given the multitasking nature of UAV missions, the concern is that achieving these goals could create serious workload problems.
In order to establish the feasibility of this goal, two approaches--pilot-in-the-loop simulation and computational modeling--should be taken in parallel. The first approach is to evaluate ways of reducing unacceptable workload levels that a 1:1 and, in particular, a 1:2 operator-to-aircraft ratio will impose on the single operator. In our UAV simulation, crafted from interviews with subject-matter experts (U.S. Army UAV pilots attached to E Company, 305th Military Intelligence Battalion), the control of each of two UAV workstations involves three major visually distributed subtasks: (a) mission completion, which involves navigating the UAV between command targets and reporting information at those locations; (b) monitoring the health of various on-board system parameters in order to detect and respond to periodic system failures; and (c) surveillance of the ground beneath each path through a 3-D camera image to find well-camouflaged "targets of opportunity" (TOOs) and then identify them through demanding image manipulation (Gugerty & Brooks, 2001). In a sense, Task a is the primary mission task, Task b supports the mission completion, and Task c can be represented as a valuable secondary task that increases the productivity of a UAV mission. In our simulation, two of these tasks are augmented by automation: an autopilot can handle all aspects of navigating the UAV (but provides no benefit to the command target report portion of this task), and an autoalert system can replace visual monitoring with auditory detection of system failures.
The second approach is to develop a computational model of multitask-processing assumptions about the pilot-UAV system in order to predict how these and other workload-relevant modifications to the system might influence pilot workload (Laughery & Corker, 1997; Waiters, Huber, French, & Barnes, 2002). Doing so can greatly reduce the time and expense of pilot-in-the-loop simulations. If the model is validated, then questions can be answered more rapidly regarding workload implications of automation and pilot-vehicle ratio modifications of the type described. The current article focuses primarily on the first approach.
A review of the literature indicates that although many human factors issues--display design, crew coordination, time delays, and so forth--have been noted in the UAV paradigm (e.g., Gawron, 1998; Mouloua et al., 2003), few of these issues have been resolved through experimental control. Previous UAV research has focused primarily on single-UAV operation (e.g., Gugerty & Brooks, 2001; Veltman & Oving, 2003), primarily via part-task simulations with few, if any, attempts to generalize to multiple-UAV operation. Other research has focused on controlling multiple robotic objects such as missiles (e.g., Cummings & Guerlain, 2003); however, these types of robotics are fairly simple to operate and make fewer demands on resources than do complex UAV systems. …