Adaptive virtual reality stress training for spaceflight emergency procedures

Abstract

Emergency training is an essential tool to mitigate safety risks to vehicles, operators, and for mission success. NASA astronauts go through extensive training to prepare for such situations. Astronauts can experience acute stress during hazardous, potentially life-threatening, situations that may erode any prior training and diminish remedial performance. Even high levels of skill training can succumb to the stress associated with the existential threat from an emergency. Incorporating stress training into the emergency training process may prepare astronauts to respond more favorably to stressful events. However, the implementation of stress training is difficult due to resource limitations, wide-variation between individual’s stress responses, optimizing training to match user competency levels, and fidelity of the training environment. The research objective is to develop and test an adaptive virtual reality (VR) stress training system as a countermeasure strategy against acute stress from spaceflight emergency operations. An adaptive VR training system may help astronauts develop resilience in preparation for high-stress operations. Four studies investigated the components and overall evaluation of the adaptive VR stress training system. The first study evaluated the effect of gradual exposure to stressors on building stress resilience. Participants were tasked with locating a fire on a virtual International Space Station (VR-ISS). Physiological and psychological measures were taken and results showed that prior exposure, as would be experienced during a gradual exposure to stress, enhanced relaxation behavior when confronted with a subsequent stressful condition. The second study developed and evaluated an emergency procedure, then manipulated a VR-ISS environment with three levels of stressors to induce psychological stress. The third study developed and tested a physiologically based stress detection system that uses personalized interval methods to classify stress levels during tasks of ever-higher complexity, including an emergency fire procedure on the VR-ISS. A classifier was developed and tested against standard machine learning classifiers. Results from a human research study show high levels of accuracy in detecting multiple stress levels, even across tasks and when compared to other machine learning classifiers. The fourth study integrated the components from prior studies and evaluated a real-time adaptive stress training system. Using a VR simulation of a spaceflight emergency fire, predictions of the individual’s stress levels were used to trigger adaptations of the environmental stressors (e.g., smoke, alarms, flashing lights), with the goal of maintaining an optimal level of stress during training. The adaptive training was compared to predetermined gradual increases in stressors (graduated), and trials with constant low-level stressors (skill-only). Results suggests that all training conditions lowered stress, but the adaptive condition was more successful decreasing multiple stress measures during the stress exposure. Lastly, the lessons learned from each of the studies was compiled into a list of recommendations to aid future researchers looking to improve training, stress detection, or adaptive systems.