S-114.220 / AS-84.J Research Seminar on Computational Science
The first topic of the course in spring 2005 is
The main topic of the course is predictive motor control in vertebrates. The goal is to learn the neuroscience, machine learning and robotics aspects of this. Motivation for this type of course is that on one hand, robotics and machine learning can learn a lot from motor control and learning in animals. Roboticists can at the moment only dream of the skillful, adaptive movements of animals. On the other hand, robotics and machine learning gives a valuable system-level view to neuroscience. Traditionally neuroscience has been dominated by analysis and experimentation but the danger is that one becomes overwhelmed by details and the big picture remains obscure.
Apart from predictive motor control, some related topics (such as saccadic eye-movement control, visuo-motor control and task-adapted sensory processing) will be covered on a theoretical level but not in robotic implementation.
Understanding the articles covered during the course will require knowledge and skills in many different disciplines (neuroscience, robotics, machine learning, programming) but the students are not required to have prior knowledge on all of these fields. Rather, the participants should be prepared to work in multidisciplinary teams and share their knowledge with other team members with different backgrounds.
|25.1||Harri Valpola||Overview of the course and introduction to vertebrate motor system|
|25.1.||Mika Vainio||Introduction to robotics|
|1.2.||Selection of teams and outline of projects|
|1.2.||Jari Saarinen (Team 1)||Basics of feedback control|
|8.2.||Team 2||Feedback control and predictions|
|15.2.||Team 3||Representations and sensory processing|
|22.2.||Team 4||Reinforcement learning and action selection|
|1.3.||Discussion about the communication architecture, first experiments and tasks|
|8.3.||Discussion about the communication protocol and details of implementation|
|15.3.||More discussion about implementation|
|22.3.||Discussion about uses of predictor: forward and inverse models|
|5.4.-3.5.||Discussion about the implementations. Each group reports their progress. Gradually the work is integrated together.|
|10.5.||Final meeting and demonstration of the integrated control architecture|
|D.Sc. Harri Valpola |
Harri.Valpola at tkk.fi