[Seminar] "Action Prediction and Object Bootstrapping: A Neurocognitive View onto a Robotics Problem"by Prof. Dr. Florentin Wörgötter

Prof. Dr. Florentin Wörgötter

Georg-August-Universität Göttingen
Bernstein Center for Computational Neuroscience 
Department for Computational Neuroscience

Title:
Action Prediction and Object Bootstrapping: A Neurocognitive View onto a Robotics Problem

Abstract:
Humans are able to perform a wide variety of complex actions manipulating a very large number of objects. We can make predictions on the outcome of our actions and on how to use different objects. Hence, we have excellent action&object understanding. Artificial agents, on the other hand, still miserably fail in this respect. It is particularly puzzling how inexperienced, young humans can acquire such knowledge; bootstrapped by exploration and extended by supervision. To address this issue we have structured the realm of actions and objects into dynamic representations, which allow for the easy learning of different action and object concepts. We were specifically interested in developing a framework that captures two specific human traits: (1) humans are capable of fast action prediction (for example during collaboration) and (2) we are often re-defining and ‘abusing’ objects according to need (for example using a hammer as a rake to fish for a nail box, which is out of reach). We address these aspects focusing on manipulation actions and show that with this framework robots can acquire mildly cognitive properties improving their action and object understanding.