This course concerns robots that can adjust and improve their behaviour
over time. Objectives: 1) To understand the difference between machine
learning and learning machines. 2) To equip a given robot with learning
abilities and experimentally test the resulting performance. 3) To
experience the difference between simulated and real robots. To this
end, we use Thymio robots enhanced with Raspberry Pi 'brains', cameras,
and powerful batteries. For the simulations we use our own software that
allows easy portability of code between simulated and real Thymios.
The course has a strong hands-on flavour. After two introductory
lectures students have to develop and implement the learning method of
their choice in simulation. In particular, adequate robot controllers
have to be learned autonomously for two tasks, maze navigation and food
collection. After testing and tuning the methods in simulation, the best
learned robot controller must be ported to a real Thymio and the real
world performance compared with that observed in simulation.
Students will work in teams of three, attending two obligatory workshops
each week. Grading is based on 1) task performance of the robots, 2)
short report of the approach and experimental results. During the last
lecture we will have a live demonstration of all robots.
Maximum number of students: 30.
Only VU-students from the AI-master (SAC or Cog.Science) can enter this