Page 48 - Innovations in Intelligent Machines

36 M.L. Cummings et al.
5 Conclusions

With the recognition that intelligent autonomy could allow a single opera-
tor to control multiple vehicles (including air, ground, and water), instead of
the converse which is true today, there is increasing interest in predicting the
maximum numbers of autonomous vehicles an operator can control. A critical
system architecture question is then how many vehicles could one operator
control? While there are other methods that could be used to predict this
number (e.g., cognitive modeling which suffers from the ability to represent
highly complex systems, and simulations and experiments with advanced pro-
totypes, which suffer from exorbitant development costs), we demonstrated,
through two different methods, how this number can be estimated by consid-
ering the temporal elements of supervisory control of multiple UAVs.
In the first method, we demonstrated that past equations of fan-out omit-
ted important aspects of human interactions with multiple UAVs. We suggest
an alternative equation that captures some of these aspects using wait times.
However, these temporal approaches to measuring fan-out are limited since
these results are not explicitly linked to performance. In comparison, we used
cost-based simulation model that links operator performance to both mission
costs and complexity; however, it suffers from problematic assumptions and
an inability to highlight specific areas for design interventions.
While each method has strengths and weaknesses, they are not mutually
exclusive. The two approaches can be synergistic in that temporal data gath-
ered experimentally for initial rough estimates such as fan-out can provide
more valid simulation models. Predictions then made through optimization
simulations can be furthered refined through sensitivity analyses and appro-
priately focused human-in-the-loop experiments. In this way, effects of increas-
ing UAVs and/or system autonomy can be seen on system performance as well
as operator performance. In terms of application, this iterative approach to
predicting operator capacity would likely provide the most benefit early in
the systems engineering conceptual stages when unmanned aerial systems are
still in development and uncertainty in system parameters is high.

Acknowledgments

The research was supported by grants from Boeing Phantom Works and
Lincoln Laboratory.

References

1. J. Franke, V. Zaychik, T. Spura, and E. Alves, “Inverting the Operator/Vehicle
Ratio: Approaches to Next Generation UAV Command and Control,” presented
at Association for Unmanned Vehicle Systems International and Flight Interna-
tional, Unmanned Systems North America Baltimore, MD, 2005.