The AFM’s AI-based path detection module, which is responsible for identifying drivable paths, was tested on public roads with an intrinsically and extrinsically calibrated front camera. The path detection module uses machine learning algorithms to detect lane markings on incoming camera images in every time step. This test was done in order to evaluate the module’s ability to accurately detect the drivable paths in the vehicle’s local coordinate system.
Further, the detected paths were fused with highly accurate GPS data. While the AI-based path detection module provides local information about the immediate surroundings, GPS data gives the vehicles global position in the world. During the tests, GPS data was used to transform the detected paths from the local vehicle coordinate system into the global world coordinate system. This combination of AI-based path detection and highly accurate GPS data will form the basis for the AFM’s real-time capable AI- and optimization based path planning and path following control algorithms.