Developmental Learning using Low-Cost Mobile Manipulators

01.12.2008 - 20.06.2010
Forschungsförderungsprojekt
The nascent field of Developmental Robotics explores autonomous (i.e. little or no supervision) and continuous "life-long" learning in robots. Robots learn based on experiments which they plan and conduct on their own. They learn how the surrounding external world works, how their internal mental world works and how these worlds interact via the sensor-motor interface. In contrast to this autonomous learning, "classical" machine learning requires the learning situation as well as the desired result to be specified, which is impossible for a robot confronted with a large number of arbitrary situations in an everyday environment. Especially in the start phase of learning error rates will typically be high and learning experiments can take hours or days, which makes them tedious to supervise. For obvious safety reasons this rules out usage of platforms typically used in mobile robotics, which weigh over 100 kg and can cost well over 50000 EUR. Therefore developmental robotics so far concentrated mostly on small and simple mobile robots (e.g. just equipped with two motors and a distance sensor) or on stationary robotic arms. Recent advances in miniaturisation especially of powerful servo motors however enable construction of complex light-weight robotic arms suitable for mounting on small mobile platforms. The fact that these are obviously less accurate than their more "professional" industrial counterparts poses no problem for developmental learning. The aim of this project is to study developmental learning on small, low-cost robots equipped with manipulators. Especially we want to investigate how manipulation actions can be learned in state-of-the-art developmental learning frameworks. Furthermore we want to investigate the dynamics of the interaction of several robots learning in the same environment, e.g. how observation of an experiment performed by another learner influences the own learning.

Personen

Projektleiter_in

Institut

Grant funds

  • Hochschuljubiläumsfonds der Stadt Wien (National) Anniversary Fund for Higher Education of the City of Vienna

Forschungsschwerpunkte

  • Cognitive and adaptive Automation and Robotics: 100%

Schlagwörter

DeutschEnglisch
RobotikRobotics
Maschinelles LernenMachine Learning
Künstliche IntelligenzArtificial Intelligence

Publikationen