Adaptive Cruise Control Regulation via Driver Model
Initial situation
Adaptive Cruise Control (ACC) systems adjust vehicle speed based on preceding vehicles, current traffic signs, curves, road grip, and other factors. Additionally, in the future, a lot of data from the vehicle interior about its occupants will be available. In this project, Fraunhofer IAO and INVENSITY GmbH are investigating whether future ACC systems can also adjust the vehicle's speed and distance to the preceding vehicle based on the driver's condition (as an additional control variable).
Solution idea
In an AI approach, data from the vehicle interior are to be used to recognize driver states and map them into driver models (e.g., stress, workload, flow).,
(Attention, intentions, emotions). The AI recognizes patterns in the available interior data as well as other multidimensional data and classifies these into driver models. An ACC adjusts the speed or distance to the vehicle ahead within predefined limits to match the driver's state. This allows
A control loop is created in which an AI learns which speed, which distance, and possibly also which music, lighting scenarios, and driving dynamics are regulated to the optimal target variables for the driver's state. Using reinforcement learning, the AI regulates the driver-state-adaptive speed.
Benefit
A driver state-adaptive cruise control increases comfort and safety when driving manually, as both stress and hypovigilance (reduced
(attention due to lack of stimulation) can be reduced. The development is also transferable to automated driving and the adaptation of driving parameters to occupant models. Successful detection of driver states and their representation in driver models is also transferable to many other applications, e.g., learning, working, sports, etc.
Implementation of the AI application
The following methods are used:
- Affective Computing Recognition Technologies
- Signal processing + Machine learning (supervised and possibly unsupervised) for arousal detection and monitoring, using, for example, Support Vector Machines, Logistic Regression, and Random Forest Classification
- Reinforcement Learning for the Calibration of Driver State Adaptive AI ACC