Self-Organising Locally Interpolating Map for the control of mobile microrobots

H. Hülsen
S. Fatikow
D.T. Pham
Z. Wangb

Keywords : self-organising map, learning control, mechatronics

The paper presents a learning controller that has been designed for the control of mobile microrobots, which belongs to the class of nonlinear, time-variant systems with ambiguous inverse behaviour. The so-called Self-Organising Locally Interpolating Map (SOLIM) consists of a continuous nonlinear support vector-based map generating control commands from desired system states, and a learning algorithm that approximates the map to a smooth inverse model of the system behaviour with help of the measured system output. The learning algorithm uses a new self-organising rule that has been specifically designed to learn a controller mapping even when succeeding sensor values are correlated. Experiments show that the controller can learn to control the velocity of a mobile microrobot by only observing its behaviour.

Submitted by Helge Huelsen on Mon, 26/06/2006 - 1:28pm.
login or register to post comments |

A question

Submitted by Afshin on Thu, 13/07/2006 - 1:13pm.

Thansk for your nice and clear presentation.

A question has been arised from one of the visitor students and that is:

Why do you use a SOLIM approximation and not a simple regulator to follow the system behaviour?

Regards

Thank you for the good

Submitted by Helge Huelsen on Thu, 13/07/2006 - 3:42pm.

Thank you for the good question.

In a mobile microrobot control context the SOLIM controller maps from desired robot velocity vectors to actuation parameters such that measured velocity vector of the robot is approximately identical to the desired one.

SOLIM thus learns one smooth inverse of the system behaviour such that the combination of SOLIM and the system has an approximately linearised behaviour. The SOLIM controller and the system are then embedded into a closed loop control structure, which measures and controls the position of the microrobot platform.

The whole controller consists of three elements:
- A trajectory controller, comparing measured pose with a given trajectory (start and end pose) and calculating a local desired pose along the trajectory and onto the trajectoy
- A motion controller, calculating a desired velocity from the local desired pose (can take the robots dynamics into account)
- An actuator controller (SOLIM), mapping from a desired velocity to actuation parameters. This can be viewed as linearisation element.

Sorry that I have not described it in the paper, I have put the focus on the algorithm ...

Please ask if you have any further questions.

Helge

Self-Organising Locally Interpolating Map for the control of mobile microrobots

Comment viewing options

User login