Learning Machines

Course Objective

1) Understanding and being able to critically discuss how learning machines relates to machine learning: This includes the main aspects involved in both, as well as their implications. (Knowledge and Understanding)

2) Designing systems of robots able to adapt to solve tasks: This includes endowing a robot with learning mechanisms that make use of sensing capabilities. (Applying Knowledge and Understanding)

3) Designing and conducting experiments with simulated and real robots to solve tasks. (Applying Knowledge and Understanding)

4) Being able to analyze the results from experiments (3), deriving insights into diverse aspects related to the methodology utilized. (Judgement)

Course Content

This course concerns designing robot systems that can sense their environment, act upon it, and improve their behavior to solve tasks.

Most of the course encompasses practical sessions with only a few lectures.

Multiple tasks should be delivered, and students have the freedom to choose different learning and optimization methods to work with.

All tasks will be done in simulation AND in hardware using Robobo robots. The Robobos are wheeled robots with fixed bodies that possess infrared sensors and a camera.

Students are expected to design/optimize the controllers (brains) of the robots to solve the due tasks while utilizing the sensors as inputs.

For the simulations, a robot programming/simulation framework will be provided to students. The simulator utilized is VREP (Coppelia).

Additional Information Teaching Methods

The course is short and requires intensive work - it happens over a period of four weeks - each week contains one lecture, two practical sessions, and one presentation & evaluation session.

Attendance at all sessions is mandatory. Missing lectures and practical sessions must be justified to the teacher and might be very detrimental to the assignment of the week, while missing a presentation incurs failing the assignment of that week.

Students will work in teams.

This is a course for students willing to go the extra mile, requiring a high level of independence.

Method of Assessment

1) Task I (presentation and demo): 20/100

2) Task II (presentation and demo): 20/100

3) Task III (presentation and demo): 20/100

4) Final report: 40/100

Tasks scores range from 0 to 10, and all tasks must be above 5.5 to pass the course.

In each presentation, the group is required to disclaim each member's contribution explicitly. Additionally, each member must participate in presenting and, most importantly, in answering questions from the teacher or TA during/after the presentation. If the student does not demonstrate enough individual knowledge, their score for that assignment will be individually penalized.

The course does not offer resits, but students can reuse assignment's grades in case they fail and enrol again.

Entry Requirements

-

Additional Information Target Audience

Recommended background knowledge

• Having passed any course involving learning algorithms, e.g., Reinforcement Learning, Evolutionary Computing, Computational Intelligence, Deep Learning, etc. - the course will not teach you new algorithms and it is fundamental that you have intermediate skills with at least one learning algorithm.
• Intermediate programming skills
• Python
• Independence for self-study

Not mandatory, but useful: Experience with robotics simulators - these can be very challenging to work with.