As an intelligent robotic vehicle, the ROboMObil belongs to the class of artificial intelligence agents. When comparing the ROboMObil with conventional cars that were modified to allow for autonomous driving, two features can be pointed out: first, the high maneuverability and second the strong focus on cameras as a perception sensor. Within the VAL, the ROboMObil incorporates a hybrid autonomy architecture consisting of a (reactive) task-based part and a classic hierarchical planning part. Entering a parking space in narrow urban environments is predestined for demonstrating the high maneuverability of the ROboMObil. All three ROboMObil motion modes (longitudinal, lateral, and rotational) can be utilized for this task. An autonomous parking approach was developed and evaluated with real world tests. A basis of the ROboMObil’s artificial intelligence are vision-based control approaches that have a reactive character, providing the potential for a fast response time and increased robustness by considering the sensor characteristics explicitly. The first developed approach of this class is a direct approach for relative positioning. This method was then improved and modified for platooning of two vehicles using only a monocular camera. Moreover, another crucial functionality was developed for reactive avoidance of static and dynamic obstacles (see figure below).
By clustering and analyzing the optical flow, this approach is able to identify potential collisions with dynamic obstacles. Epipolar geometry is exploited to derive velocity commands that ensure a collision free path for the ROboMObil via a real-time optimizer. This reactive algorithm works in the abstract representation of image space instead of a scale-true representation, which provides a faster response time and less accumulation of errors. The transition from the image space to the scale-true/metric space is only made when the optimizer outputs the correction.