Cognitive Robotics and Learning
The research group is organized at the border of machine learning, cognition and human-robot interaction. The ultimate goal is to enable interactive learning in human-machine cooperation. The research group investigates efficient and life-long learning methods for behavior generation, motor learning and visual object recognition to maximize the autonomy and adaptability of robotic systems. In particular, imitation learning and bootstrapping processes are central research areas. Application areas of this research group comprise behavior learning and generation for ASIMO, iCub, the light-weight KUKA LWR robot arm and other robots, visual online learning for object recognition, and autonomous learning approaches to generate complex time series including audio and video signals.
Main methodological foci are neural learning methods, in particular recurrent reservoir networks, and the transfer of other machine learning approaches to interactive scenarios, which require high computational efficiency and online-learning capabilities. Of particular interest in this domain are generative approaches to allow for behavior generation along with classification or prediction. Important methodological aspects of generative models are covered by the projects "ALEGRO" and "Theory of generative reservoirs".
The projects "Neural Learning of Flexible Full Body Motion" and "Goal-directed Imitation Learning from Humans" pursue the autonomous bootstrapping of motor control and skill acquisition by imitation learning. Central goal is to understand cognitive and developmental aspects related to motor learning and learning by imitation. The research group follows a synthetic approach to tackle important questions in this context by implementing computational models of these processes.
The research group also contributes neural learning methods and robotic experimentation to the FP7-IP large scale project AMARSi -- Adaptive Modular Architectures for Rich Motor Skills, which is coordinated by Prof. Steil.
(Only projects directly associated with this CoR-Lab research group are listed, for further projects of J. J. Steil see here)