Tamim Asfour - Engeneering Humanoid Robots that Learn from Humans and Sesorimotor Experience

CITEC Summer School 2017 on Cognitive Architectures.

Keynote by Prof. Dr.-Ing. Tamim Asfour

Engineering Humanoid Robots that Learn from Humans and Sensorimotor Experience

Humanoid robotics has made significant progress and will continue to play central role in robotics research and many applications of the 21st century. Engineering complete humanoid robots, which are able to learn from human observation and sensorimotor experience, to predict the consequences of actions and exploit the interaction with the world to extend their cognitive horizon remains a research grand challenge. In this talk, I will present recent results and discuss future directions of research, which combine robotics, machine learning and computer vision to speed-up learning and facilitate intuitive programming of complex robotics tasks. I will discuss how a robot motion alphabet and a robot internet of skills can be learned from human observation and how statistical language models, whose words are robot poses, and whose sentences represent sequences of poses can be learned from human motion data and used to generate complex robot motions. I will introduce the concept of “Object-Action Complexes”, a co-joint representation of perception-action dependencies, which emphasizes that objects and actions are tightly coupled, intertwined and even equivalent. Based on such representations, generative mechanisms can be defined that uses existing robot experience together with new observations to supplement the robot's knowledge with missing information about, object-, actionas well as planning-relevant entities leading to improved learning and prediction capabilities.

Discussion moderated by Dr.-Ing. Sebastian Wrede Cognitive Systems Engineering