Labyrinth 1
Development of Learning Architectures and Experiments in Sensory Motor Learning
The "Labyrinth 1"-Project uses an altered BRIO labyrinth to investigate by means of EEG and fMRI examinations brain processes that occur during learning and relearning. On the other hand, an artificial agent is supposed to be able to play this game. Currently, this game may be controlled by hand, directly via motors, or via motors and joystick. The integration of a number of sensors (potentiometers, piezo sensors, cameras, and switches) makes it possible to record the behaviour of the player as well as that of the artificial agent and to compare these with each other. In addition, a physical simulation was written in order to simulate the playing and learning behaviour of the artificial agent.
| Duration: | 15.06.2007 till 31.12.2007 |
| Donee: | German Research Center for Artificial Intelligence GmbH |
| Application Field: | Assistance- and Rehabilitation Systems |
| Related Projects: |
Labyrinth 2
Setup as testbed for learning architectures and EEG/ fMRI- analysis
(01.2008-
01.2009)
|
| Related Robots: |
BRIO Labyrinth
Testbed for the development of learning architectures
|
| Related Software: |
MMLF
Maja Machine Learning Framework
|
Project details
The aim of the project is the integration of servo motors into the partially rebuilt BRIO Labyrinth and to optimize video recording. With the aid of servo motors it will be possible to change the spatial position of the playing area in two layers. Except for a load test, the servo motors to be used have been tested and approved. However, it is not expected that these servo motors will be exposed to heavy loads in our planned applications. A programme will be written to enable a controlled drive of the servo motors. Apart from an analysis of potentiometer data, the evaluation of visual data will support the analysis of the movement on the playing area and of the ball in the labyrinth. In this project, video recording will be optimized and a programme for the input of video data will be written. Upon finalizing the last steps of integration, the game will be used for EEG and FMRI studies and as a platform to implement and to test different learning architectures as well as evolutionary algorithms. The reconstructed BRIO game is considered a prototype which will be improved and duplicated once tests have proven successful.
Current state of the game:
- The game was altered and is more or less unmagnetic; thus, it may be employed for fMRI investigation in the tomograph.
- Two high-resolution potentiometers are integrated to measure the position of the axes. Data may be recorded via a Labview-based programme (author: José de Gea) and employed to describe the behaviour of the test persons.
- A LED light integrated at the edge of the game may send light impulses and thus help to synchronize the data from potentiometers and videos as well as from recorded fMRI and EEG data, resp. The LED light is also controlled by the Labview programme.
- A piezo sensor was integrated that registers the vibrations of the game board when the ball falls through a hole. These are being recorded via the Labview programme.
- Both rotating disks of the game will be substituted by two disks which may be operated by persons or connected to motors.
- Fixtures for DC motors were made, however, these are difficult to handle and should not be employed, particularly, since operation has switched to servo motors
Videos
BRIO Labyrinth: Development of learning architectures and experiments in sensory motor learning
The BRIO labyrinth game is used as a testbed for neurobiological studies of sensorimotori integration.
BRIO Labyrinth: Learned Policy
Simulation of the learned policy at the BRIO Labyrinth.