Final Report Summary - ISI-PADAS (Integrated human modelling and simulation to support human error risk analysis of partially autonomous driver assistance systems)
Executive summary:
The main objective of the ISI-PADAS project was to improve the current design process of driver assistance systems, such as Partially autonomous driver assistance systems (PADAS), by effectively introducing a tool-supported risk based design methodology for the evaluation of hazards associated with human errors and inadequate driver behaviour. The basic goal of the improved methodology for Risk-based design (RBD) was to be able to create predictions of critical or error-prone situations for drivers, by a modelling and simulation approach. The basic idea is to do a huge amount of fully automatic simulations, based on models of the vehicle, the environment, the PADAS and the driver. The data from these simulations are used to support the risk assessment for new systems. The major advantage of this approach is the substantial gain in speed of evaluation.
Heading in that direction ISI-PADAS developed a Joint driver vehicle environment (JDVE) simulation platform. The JDVE serves as the core technical infrastructure of the simulation process and as a flexible framework for the integration of different models and tools. Throughout the project the JDVE was used in different configurations - using human subjects or driver models; using PADAS or no assistance; and collecting data from real or simulated cars.
Several studies in different scenarios were carried out within the project. One set of studies focused on analysing driver behaviour without a PADAS, whereas another set investigated driver behaviour in interaction with two different PADAS for the support of longitudinal control. Both PADAS variants were developed within ISI-PADAS and were connected to the JDVE as software prototypes.
Based on the knowledge gained from the studies a set of five heterogeneous driver models were developed. One of these models is a distraction classifier, which is able to detect a distracted state of the driver without relying on additional sensors. It is used to support the PADAS system by providing information about the driver. Two other models are simulation models of the driver. These were connected to the JDVE as a replacement for human drivers. They proved to be able generating relevant aspects of human driver behaviour. In a selected use case the driver models were used to simulate a scenario 10 000 times while interacting with a PADAS and 10 000 times while having no PADAS integrated.
Critical situations are most relevant for the risk assessment process but do not occur very often even for comparable large number of simulations. To provide a more sophisticated estimation of probabilities of critical but rare situations an approach based on extreme value theory was used.
To handle all these processes a support tool for the risk analyst was created. The user can assess the risk by decomposing all possible scenarios starting from some selected initial situations. The tool supports the process by providing Expanded human performance event trees (EHPET) as structural component for the decomposition of scenarios.
Finally, the applicability of the improved RBD methodology was successfully demonstrated on traffic light approaches as use case scenario.
The main objective of the ISI-PADAS project was to improve the current design process of driver assistance systems, such as Partially autonomous driver assistance systems (PADAS), by effectively introducing a tool-supported risk based design methodology for the evaluation of hazards associated with human errors and inadequate driver behaviour. The basic goal of the improved methodology for Risk-based design (RBD) was to be able to create predictions of critical or error-prone situations for drivers, by a modelling and simulation approach. The basic idea is to do a huge amount of fully automatic simulations, based on models of the vehicle, the environment, the PADAS and the driver. The data from these simulations are used to support the risk assessment for new systems. The major advantage of this approach is the substantial gain in speed of evaluation.
Heading in that direction ISI-PADAS developed a Joint driver vehicle environment (JDVE) simulation platform. The JDVE serves as the core technical infrastructure of the simulation process and as a flexible framework for the integration of different models and tools. Throughout the project the JDVE was used in different configurations - using human subjects or driver models; using PADAS or no assistance; and collecting data from real or simulated cars.
Several studies in different scenarios were carried out within the project. One set of studies focused on analysing driver behaviour without a PADAS, whereas another set investigated driver behaviour in interaction with two different PADAS for the support of longitudinal control. Both PADAS variants were developed within ISI-PADAS and were connected to the JDVE as software prototypes.
Based on the knowledge gained from the studies a set of five heterogeneous driver models were developed. One of these models is a distraction classifier, which is able to detect a distracted state of the driver without relying on additional sensors. It is used to support the PADAS system by providing information about the driver. Two other models are simulation models of the driver. These were connected to the JDVE as a replacement for human drivers. They proved to be able generating relevant aspects of human driver behaviour. In a selected use case the driver models were used to simulate a scenario 10 000 times while interacting with a PADAS and 10 000 times while having no PADAS integrated.
Critical situations are most relevant for the risk assessment process but do not occur very often even for comparable large number of simulations. To provide a more sophisticated estimation of probabilities of critical but rare situations an approach based on extreme value theory was used.
To handle all these processes a support tool for the risk analyst was created. The user can assess the risk by decomposing all possible scenarios starting from some selected initial situations. The tool supports the process by providing Expanded human performance event trees (EHPET) as structural component for the decomposition of scenarios.
Finally, the applicability of the improved RBD methodology was successfully demonstrated on traffic light approaches as use case scenario.