Highly automated driving functions allow people to be completely relieved of driving tasks in defined scenarios. The required algorithms for environment perception based on sensor data are already at a good level today. The great challenge is the scene understanding and predicting the behavior of other road users which is subject to uncertainty. These factors form the basis for decision making and planning the route and trajectory of the ego-vehicle. Last but not least, the behavior of the vehicles interact with each other.
The research project's objective is to develop a function that builds an comprehensive understanding of the environmental condition and predicts the behavior and trajectory of other road users. The focus is on surrounding areas such as container stations, mines and construction sites. The function to be developed should explicitly take into account the information from the perception of the environment and the interaction with other road users. From the fields of rule-based, model-based and data-driven methods, the one with the best suitability for the realization of the target function is to be selected. On the basis of the prototypical implementation, the application spectrum of possible scenarios for the application of the function should be evaluated.
As a basis a broad catalogue of the requirements, the state of the art of current prediction approaches and the benefit of the function has to be evaluated. Subsequently, existing data sets will be extended with specific commercial vehicle data and prepared for the training of data-based methods. The next steps are the interface specification to derive the software architecture of the target function and the setup of a simulative development environment. In this environment the algorithm is iteratively developed by using and comparing promising methods. Finally, the predictive function will be validated in the vehicle prototype using predefined evaluation criteria.