Abstract:

Automated parking is an interesting and challenging research and development area, both for academia and commercial companies, in the context of intelligent transportation systems. The implementation of such an automated parking system requires interdisciplinary knowledge of both control theory/engineeering and of artificial intelligence/computer vision methods. Various types of sensors and control programs have been used for achieving the goals of high accuracy, functional safety, and passenger comfort.

Lately, deep learning also has been involved in the development of self-driving and self-parking cars, in areas like image recognition, modeling of car-following and lane-changing behaviors, driving behavior analysis, learning interactive driving patterns in intersections, driving maneuver early detection.

Besides the hardware and software capabilities of such cars that are able to park by themselves, on a higher level there are the parking strategies which are an area of intense studies in transportation engineering. For example, smart parking in a IoT-enabled city is an area that draws important efforts in the context of more sustainable citities. Parking strategies should assist and inform drivers in finding and choosing the optimal parking place.

In this paper, we propose the implementation of simple parking strategies (meek, prudent, and optimistic) for a parking lot modeled as a semi-infinite line where cars enter from the right at a fixed rate, and where the spatial distribution of cars depend on the parking strategy. These strategies are modelled in the literature and are the building blocks of our implementation. Based on these strategies, we demonstrate these strategies using in-house built autonomous cars (car-like robots). Thier abilities are based on a wide range of sensors and control programs based on artifical vision and deep learning, that allow them to successfully park given one of the three above mentioned parking scenarios.

The system as a whole can be used for teaching modeling and simulation introductory concepts (modeling mechanical systems and robots), control engineering (basic control laws), artifical intelligence (computer vision), and deep learning algorithms (reinforcement learning). Based on the system physical and software components, we describe in full details different learning strategies and the benefits for the learner.

Keywords:

Control theory, artificial intelligence, self-driving cars, sensors, computer vision, control laws, reinforcement learning.