Simulation and Classification of Spatial Disorientation in a Flight use-case using Vestibular Stimulation

In aeronautics, spatial disorientation (SD) is "an erroneous sense of one’s position and motion relative to the plane of the earth’s surface" (Gillingham, 1993). SD has a wide range of situations and factors, but mainly it has been studied using reduced experimental contexts such as motion detection experimentation in isolation. Because there are many SD use-cases that are studied in isolation in a reduced manner, it is difficult to develop a generalized and fundamental understanding of the occurrence of SD and viable solutions. To investigate SD in a generalized manner, a two-part neuroergonomics study consisting of an in-flight piloting use-case experiment and machine learning (ML) model prediction was performed. The first part of the study was the creation of a generalized SD perception dataset using whole-body experimental motion detection methods in a naturalistic flight context; participant perceptual joystick response was measured during rotational or translational vestibular stimulation. The second part of the study consisted of ML parameter tuning selection for SD prediction, using joystick response-derived features from the generalized SD perception dataset. Additional measures of SD were investigated for future ML feature usage, such as questionnaire-based physical disorientation measured using the simulator sickness questionnaire (SSQ) disorientation sub-scale. The perceptual SD dataset was statistically proven to be representative of human motion detection behavior, demonstrating that the simulation environment was sufficient to generate a fidel SD context. ML modeling comparison analysis demonstrated that SD can be accurately predicted regardless of the feature quantity used, however model type, specialized dataset models, feature type, and label type significantly influence prediction accuracy. Finally, no significant relationship between physical disorientation and motion detection was found, indicating that two-sample before and after SSQ questionnaire-based methods are insufficient to uncover correlations with perceptual disorientation; a more frequent physical disorientation measure is needed.