Assessing The Impact Of Autonomy And Overconfidence In Uav First-Person View Training

With the rapid rise in unmanned aerial vehicles (UAVs) for military and civil first-person applications like infrastructure inspection, there is an increased need for skilled UAV operators. However, research on effective training of UAV pilots has not kept pace with the demand. How much autonomy should be onboard, how much training, and how much control humans should have are still points of debate. To help fill this gap, this paper examines how different training programs and levels of control autonomy affect training outcomes for people operating a UAV in inspection tasks with high onboard autonomy. Results revealed a cost-benefit trade space in that those top performers with both lower-level teleoperation and higher-level supervisory control training could achieve the best performance, but with higher variability, as compare to those who received just supervisory control training. Another important finding was that those trainees who were overconfident were more likely to spend too much time micro-controlling the UAV, and also 15 times more likely to crash. Given that commercial UAV licensing is expected to significantly increase in the next few years, these results suggest more work is needed to determine how to mitigate overconfidence bias both through training and design.