Technische Fakultät, AG Neuroinformatik
Universität Bielefeld
Abstract
Grasping intelligence has evolved to allow complex, multifingered hands to control sophisticated
mechanical interactions, enabling tool use and preparing higher forms of cognition. Therefore,
replicating grasping intelligence for robots may give us valuable clues for grasping intelligence
at the levels of thinking and communication. Starting from an account of elements that have to be
combined in order to realize grasp intelligence, we present an architecture specifically developed
to deal with the close intertwining of subsymbolic, continuous control and a level of more
symbol-like, discrete decisions. We then show how results achieved in the context of projects in
SFB 360 tie together to enable and guide highly varied grasps with multifingered robot hands,
including examples of vision-based guidance of grasp types, the combination of internal,
physics-based si-mulation and imitation to constrain grasps and the use of concept-driven grasp
strategies to achie-ve robust grasping of complex objects. We finally conclude by discussing
bridges that may lead us from grasping intelligence towards a better grasping of intelligence.