Advanced Topics in Artificial Intelligence: AI for the Physical World

COMP 4620/8620 – Semester 2 2024#

Lecture: Mondays 12-2pm in the G6 Psychology Lecture Theatre (Building 39)

Artificial Intelligence (AI) has become a household name and is now part of our daily lives. But most of it happens in our phones or online. There is not much AI out there in the real world, at least nothing as sophisticated as what’s online. For example, there are no general purpose autonomous robots that can do things as good as or better than humans, no AI that can seamlessly interact with the real world. Such AI has the potential to make society safer and more efficient, from reducing the need for humans to perform dangerous jobs to enhancing services in remote areas to helping busy parents with everyday chores.

The use of AI in the physical world is still very limited due to several reasons. One reason is that the real world is continuous, for every action, such as “pick up” there are almost unlimited ways of executing that action, and the consequences of actions are not exactly known. This is related to the uncertainty that is ubiquitous in the real world: the exact location or physical properties of entities are unknown, perception and background knowledge is not perfect and incomplete. The real world is dynamic and changes, we often encounter new things we didn’t expect, don’t know, haven’t learned or haven’t been trained for. Humans are experienced in adapting quickly to these situations, state of the art AI is not. There are many challenges that need to be solved before we can have the physical AI we would like.

The goal of this class is to prepare students to become part of this challenge, to actively learn the skills required and to become familiar with the different approaches under consideration for solving this challenge. This is a hands on class, very different from the typical classes being offered and requires active participation from students. Students are required to read, review, present and discuss research papers on the different topics. In addition, students will work in teams on a small research project to familiarise themselves with these challenges. Active participation and attendance is required for several classes and tutorials. Some assignments require you to be present in class and the tutorials.

Due to the hands on nature and active participation requirements of this class, the student number for this class is limited. If you cannot attend classes and tutorials regularly, or if you prefer a lecture based (passive) course over a research-based course with active participation and attendance, we recommend you not to enrol this class.

Each class will be dedicated to one topic and we will prepare research papers for each class. Every student needs to select one paper in week 1 and present this paper in the corresponding class. In addition, students need to select three more papers, review them and contribute to the discussions when the paper is presented by a fellow student. The papers will be assigned to the students who have not selected the presentation paper or enough review papers. Students are also asked to give feedback and answer questions about the presentations. See the course outline for more information about the course structure.

Summary of what you need to know:

• When: Mondays 12-2pm (Canberra timezone)
• Where: G6 Psychology Lecture Theatre (Building 39)
• Lecturer: Prof. Jochen Renz, Dr. Peng Zhang
• Pre-requisites:
  • for COMP4620: COMP3620
  • for COMP8620: COMP6320
• Evaluation:
  • 3 Presentation Assignments (50%)
    • Paper presentation and discussion (20%)
    • 3 paper reviews (15%)
    • 6 presentation evaluations and mini quizzes (15%)
  • 7 Project Assignments (50%)
  • Assignments have 100% penalty for late submission
• Passing grade: 50% or more