Our group works on understanding how humans use their visual system actively during natural extended behavior to guide decisions and actions with their bodies. This leads to the study of how the visual system uses sensory input, forms beliefs about the world by carrying out computations on the basis of its representations, and then carries out actions in goal directed behavior. Both experimental evidence from the area of active vision as well as theoretical advances in control theory and machine learning show that perception, action, and learning can only be separated under very restricted circumstances. Our work therefore emphasizes the necessity to consider perception, cognition, decision making and actions jointly and investigates their interplay.
To make progress in this area we use an integrated approach requiring us to apply different methodological techniques. Our current focus is on:
- behavioral studies involving eye tracking of human eye movements during complex naturalistic tasks in natural and virtual environments,
- building models of tasks and developing algorithms that learn how to solve these tasks in virtual agents,
- developing models for the representation and quantification of extended sequential human behavior,
- simulation of learning algorithms in naturalistic virtual environments,
- developing learning algorithms with an emphasis on the learning of sensory representations for actions.
Our current collaborators include:
Christos Dimitrakakis, EPFL
Dana Ballard, University of Texas
Jochen Triesch, Fias
Jozsef Fiser, Brandeis University
Mary Hayhoe, University of Texas
Frank Jäkel, University of Osnabrück
Paul Schrater, University of Minnesota
Ralf Galuske, Technical University Darmstadt