Individualized Fatigue Management Program Technology for Trucking Operations

Individual differences in vulnerability to sleep loss and fatigue from extended work hours and night work are a substantial problem in transportation policy making, work schedule development, fatigue risk management strategies, and prediction of performance impairment in real-world operations (Van Dongen et al. 2003; Mollicone et al. 2010). Our group was the first to document the considerable magnitude of individual differences in vulnerability to the adverse consequences of fatigue (Van Dongen et al. 2004). We also demonstrated that individual differences are a biological trait, which means they are predictable (Van Dongen et al. 2005). Moreover, we showed that individual differences are as relevant in operational settings as they are in the laboratory (Van Dongen et al. 2006). Subsequently, we were the first to develop a mathematical methodology to predict performance deficits due to fatigue in individual subjects (Van Dongen et al. 2007). The overarching objective of this project (through Phase II) is to translate this basic and applied research into an Individualized Fatigue Management Program (FMP) system for the commercial trucking industry that accounts for individual differences in vulnerability to fatigue stressors to improve schedule development and manage driver fatigue and crash risk. At the heart of our approach to achieve an individualized FMP tool for trucking is a state-of-the-art alertness prediction model, which we will individualize using our proprietary closed-loop, Kalman filterbased algorithms (Mollicone 2009; Mott et al. 2009). This approach is based on our previously published Bayesian approach (Van Dongen et al. 2007). The individualized alertness model is used to predict future performance of a given individual. Simultaneously, the current, actual level of performance of the individual will be measured by capturing data from the trucking instrumentation. Through comparison of the model predictions with actual observations and by use of Bayesian optimization, the parameters of the alertness model will be updated so it is optimally tailored for the individual at hand. The alertness model will also take into account estimates of sleep/wake history from wrist actigraphy (a validated tool for estimating sleep time) (Kerkhof and Van Dongen 1996). Figure 1 provides an overview of our approach that involves collecting and integrating two primary data streams: (1) recent driver sleep patterns (e.g.,from actigraphy or sleep diaries); and (2) a real-time metric validated to track the biological fatigue status of the driver (e.g., PERCLOS) to individualize fatigue traits. Driver fatigue traits are instantiated in the system as individual-specific, fatigue-related model parameters such as “sensitivity to night driving”, or “sensitivity to long periods of wakefulness”. The primary data stream is augmented with one or more secondary data streams to additionally track the effect of driver fatigue on vehicle performance (e.g., lane tracking accuracy). The secondary data streams helps to achieve a robust system that captures interindividual differences in how fatigue affects actual vehicle performance (not just driver fatigue) and continually “self-validates” by tracking the extent to which measures of driver fatigue and vehicle performance are convergent. Through this process, the FMP tool is able to generate estimates about current and future fatigue and driver performance levels to achieve a system that can be used to improve schedule development and route planning, assist in the delivery of fatigue countermeasures, and be uplinked to headquarters for telemonitoring.