In the AUTOtech.agil project, various use cases for automated driving at low speeds (e.g. automated valet parking, traffic-calmed areas) are to be realized through the architecture of a cross-application driving function (“low-speed function”). For this purpose, a higher-level module for planning a global target trajectory from a starting point to a destination point for unstructured and semi-structured environments (e.g. parking lot) has already been developed and integrated into a simulation framework based on IPG CarMaker and ROS2. The global trajectory planner only considers static obstacles from a previously known high-definition map. Based on this, a local trajectory planner is to be developed in this master's thesis, which safeguards the specified global trajectory by reacting to static and dynamic objects in the environment with suitable maneuvers (braking, swerving, etc.). The aim is also to locally optimize the global trajectory so that it can be processed by a downstream controller. The trajectory planner is to be implemented in C++ in accordance with the interfaces of the simulation framework (CarMaker, ROS2).

In the UNICARagil project, a modular platform for automated driving was developed. With the follow-up project AUTOtech.agil we focus on the quality of data and processes. Regarding perception, the focus changes to potential savings in energy consumption and latency without losing perception quality. Therefore, we aim at methods for a requirement-driven selection of artificial neural networks (ANNs) and a review of approaches on how to optimize them.