Modelling driver behaviour in run-off-road crashes: Applications in safety system development and safety benefit estimation
Run-off-road crashes have been identified as a major concern for automobile safety and several advanced driver assistance systems (ADASs) targeting run-off-road crashes are on the market today. Assessment of ADASs require relevant test scenarios and valid computational models of driver behaviour. Therefore, the objectives of this thesis has been: (A) define run-off-road test scenarios, and (B) identify a conceptual framework suitable for modelling relevant behavioural mechanisms for crash causation.
Cluster analysis was applied to run-off-road crashes from representative in-depth crash data from the German GIDAS database. Nine different clusters were identified, forming a basis for test scenarios. The two largest clusters included crashes relevant for current lane support ADASs (i.e. drift during daytime/night-time), while other clusters suggested that drivers may need support in judging the physical limits of the vehicle (e.g. on snowy rural roads). However, a need for more detailed driver behaviour data was identified. Indeed, naturalistic data, which include more information about driver behaviour in critical situations, may help the definition of test scenarios by linking them to the behavioural mechanisms contributing to the crash causation.
This thesis also shows that modelling of driver behaviour may be supported by a framework based on new findings in contemporary neurocognitive science and, specifically, on predictive processing. This new framework improved the interpretation of the clusters and facilitated the formulation of plausible behavioural mechanisms leading to run-off-road crashes.
advanced driver assistance systems