Bayesian Network Model of Pilot Response to Collision Avoidance System Resolution Advisories

Abstract

Listen

Translate article

The effectiveness of an airborne collision avoidance system is influenced by the manner in which pilots respond to the system’s advisories. Current pilot response models used in collision avoidance system modeling and simulation are agnostic to properties of individual encounters affecting pilot response, such as encounter geometry and system alerting behavior. Therefore, the pilot response behavior described by these models does not vary between encounters. Simulations using these models can lead to inaccurate results, potentially including the underestimation of collision risk in encounters where pilot response probability is low. This paper proposes a method to construct parametric pilot response models in which pilot response to collision avoidance system advisories varies based on individual encounter properties. A model was constructed from radar recordings of Traffic Alert and Collision Avoidance System (TCAS) encounters. The encounter properties with the strongest influence on pilot response were identified using a Bayesian network. The identified properties were used to predict the probability that pilots would comply with TCAS resolution advisories in individual encounters. The model was then employed in simulation of safety-critical encounters. The same encounters were also simulated using a nonparametric model in which pilot response probability did not vary between encounters and was instead equal to the average response probability predicted by the parametric model. Results showed that the nonparametric model underestimated collision risk relative to the parametric model. This study has implications for the design and performance evaluation of separation advisory systems, including collision avoidance and detect and avoid systems.