Objectives and Contextualisation

Autonomous navigation refers here to the scientific and technical knowledge associated with the intelligent movement of physical systems. This broad definition encompasses a variety of systems, including small objects moving through the bloodstream to monitor and ensure health, household or industrial robots aiding in tasks requiring mobility, unmanned aerial vehicles monitoring forest areas to prevent fires or perform rescues, unmanned underwater robots exploring the seafloor, autonomous vehicles for delivering goods, and robotaxis or autonomous buses for urban transportation. Each of these systems presents its own unique characteristics and primary challenges. In fact, to capture the main commonalities and differences between multiple systems with autonomous navigation capabilities, we could propose several separate subjects such as body nanobots, underwater autonomous navigation, air autonomous navigation, unstructured land autonomous navigation, and structured land autonomous navigation.

Certainly, all of this knowledge cannot be covered by a single 6-ECTS university subject. Therefore, it is not surprising that, when it was planned as part of the UAB's AI degree, the focus was on autonomous driving technology for vehicles such as cars, trucks, or buses that must transport people and goods in structured land environments. Still, autonomous driving is based on multi-disciplinary knowledge, thus, we have to emphasize certain contents over others depending on the context of the subject. For instance, in the context of a degree in Telecommunications Engineering, Vehicle-to-X cooperative driving would be especially relevant; in the context of a degree in Automation and Industrial Systems Engineering the topics of local planning and control within a classical autonomous driving stack would be core knowledge; in the context of a degree in Physics, the sensors enabling autonomous driving may deserve especial interest. The subject that brings us together is within the UAB's degree in AI. Therefore, we willfocus on scientific and technical knowledge at the intersection of AI and autonomous driving (AD).

As a matter of fact, this field is still broad and continuously evolving, and there will be many topics not covered. We prefer to focus on fewer topics in more depth rather than covering more topics superficially. Considering the experience in this subject gathered in previous courses, we will focus on sensorimotor deep models for autonomous driving (sometimes referred to in the literature as end-to-end AD), where imitation learning plays a key role. Note that this is the “purest” AI and naturalistic paradigm. Therefore, we leave out traditional AD stacks based on explicit modules for semantic perception, local planning and control, HD 3D maps and SLAM, since these stacks arise from the divide-and-conquer approach of classical engineering. Moreover, covering all the paradigms would be only possible in a purely theoretical approach to the subject, while we want practical work to be core.

The teaching methodology will follow a project-based approach. We will review the basic theoretical concepts so that students can grasp the overall picture and delve into the specific aspects needed to solve a project on autonomous driving. The specific project to be solved will depend on the particular computational resources and budget for materials that are assigned to our subject (these do not depend on the teachers and may change every year). However, in general terms, we aim to follow the usual approach we can find in the real world, which involves simulation and real-world developments.

Overall, the objective of this subject is to expose the students, through theory but mainly practice, to the relevant core contents of AD, so that in the future they are ready to tackle new challenges in autonomous navigation. Last but not least, it is crucial to learn how to collaborate in a team to achieve project success effectively.