One of the most important metrics with respect to safety while driving are the mental or cognitive resources available to the driver. If these resources are depleted, reaction times increase, dangers are overlooked, and warnings are missed, even though the brain still applies a specific order. A high cognitive load also leads to stress and other physical reactions, and can even decrease steering performance. Contrary, a low resource load over a longer period of time can lead to loss of attention too.
Considering cognitive resources matters to us as researchers for two reasons:
- Each new assistance system (ADAS), apart from the expected driver support other improvements, also means an added cognitive load (depending on type and degree of interaction with the driver). Keeping an eye on these effects and quantifying them is an important task.
- Knowledge about the current cognitive load of the driver can be utilized to actively counteract an over- or under-load. This can be achieved by dynamically adjusting the behavior of all ADAS, in particular those with HMI.
Cognitive load cannot be measured directly. However, there are the following basic approaches to estimating it:
- Effects of high cognitive load can be measured. Indications of stress can be measured e.g. via biosensors (increased heart rate, skin conductance) or deterioration of driving performance (deviation from an ideal line as integrated in the driving simulator OpenDS)
- Predicting effects of interactions and stimuli on the cognitive load. Given the knowledge about the user’s behavior and the influences of the environment, based on statistical methods, the relative implications can be estimated. This assumes that the corresponding models are available.
The latter method is currently one of our main research topics, since it is much more suitable for realization within the driving car. A cognitive model of the driver, as well as a reference catalog of interaction costs, are to be integrated into our dialogue platform SiAM-dp in order to adjust the dialog flow to the user state as described above on the one hand, and on the other hand provide a function for automatic offline evaluation of dialog models.