In-vehicle pre-crash safety systems have been used by the automotive industry for some time as integral parts of advanced driver assistance systems (ADAS) and automated driving systems (ADS), in order to improve traffic safety worldwide. Among the methods used to assess these systems, virtual safety assessment methods have been shown to have great potential and efficiency. These methods are likely to continue to play a very important role in assessing vehicle safety at all levels of automation. The ultimate aim of this thesis is to enhance the safety performance of conflict and crash avoidance systems through the use of computational driver behavior models. The work first addresses this aim by incorporating behavior models into precrash safety systems; the second part focuses on overcoming methodological challenges encountered in safety system development and assessments.
The first objective of this thesis is to investigate the safety performance of safety systems that include a driver model incorporating drivers' comfortable behaviors in its crash avoidance algorithm. Chinese car-to-two-wheeler crashes were targeted; automated emergency braking (AEB) algorithms which include drivers’ comfort zone boundaries (CZB) were compared to a traditional AEB algorithm. The proposed algorithms showed larger safety performance benefits, indicating that including computational behavior models in the algorithms of pre-crash safety systems may reduce the number of crashes and injuries on our roads. It should also be noted that residual crash characteristics did not differ among different AEB implementations. If incrash protection systems do not have to account for different AEB outcomes, then the systems' designs could be simplified, leading to a more effective allocation of resources.
The second objective is to develop a method for the efficient collection of human-participant data, for use in the development of safety systems that incorporate driver behavior. The resulting method, predictive Bayesian optional stopping (pBOS), enables early stopping—either when a specific statistical target is reached or when it is not likely that the target will be reached, given the available resources (e.g., financing or test-track time). The results show that traditional Bayesian optional stopping (BOS) outperforms traditional frequentist sample size determination—and pBOS outperforms traditional BOS when the experiments have less than a 50% chance of reaching the target with the allocated resources. Consequently, under the appropriate conditions, the use of pBOS in the development of pre-crash safety systems is likely to reduce the resources required, allowing them to be reallocated to other safety research or system development priorities.
The third objective is to develop and apply a method for efficient sampling in crash causation model-based scenario generation for virtual safety assessment. The method, which is machinelearning-assisted, actively and iteratively updates the sampling probability based on new simulation results. The method requires almost 50% fewer simulations than traditional importance sampling. In addition, the impact on efficiency of incorporating the following three features into the method was investigated: domain knowledge-based adaptive sample space reduction logic, stratification, and batch size (the number of samples per iteration). The results show that both knowledge-based logic and stratification can reduce the target estimation error, and a larger batch size is preferred for overall simulation efficiency. As with pBOS, active sampling in behavior model-based pre-crash safety system assessment may reduce development costs, allowing the reallocation of resources.