Mini autonomous car architecture for urban driving scenarios

Karabulut, Gökhan
Autonomous cars capable of driving in city traffic have been long studied in architectures decomposed into perception, planning, and control components. Recent advances in deep learning techniques considerably contributed to the perception component of this approach. These techniques also laid the groundwork for the progress of other approaches such as end-to-end learning of steering commands and driving affordances from camera images. Though these approaches are promising to simplify the overall architecture, the decomposed architectures are found more persuasive, constituting the majority of today's state-of-the-art, market-oriented driverless cars. However, studies on small-scale autonomous cars, which are considered low-cost and rapid prototyping platforms, are not on a par with research on the modern decomposed architectures. These studies often remain limited to end-to-end approaches or resort to traditional image processing techniques in over-simplified traffic scenarios. In this thesis, we present a decomposed architecture for small-scale cars covering extended traffic scenarios with seven traffic signs, traffic lights, lane changes, cloverleaf interchange, pedestrian crossings, and parking. To realize this architecture, we created segmentation and classification datasets. We trained two deep learning models for learning lane semantics and classifying traffic signs and lights. We developed a behavior planner to decide on the best behavior primitives for traffic scenes. Based on these behavior primitives, we implemented a trajectory planner to find optimal trajectories along the lanes and a controller to follow these trajectories. With our novel lane segmentation scheme, 97% accurate classifier, robust planner and controller algorithms, we achieved successful drives on simulated and real courses.
Citation Formats
G. Karabulut, “Mini autonomous car architecture for urban driving scenarios,” Thesis (M.S.) -- Graduate School of Natural and Applied Sciences. Computer Engineering., 2019.