Research

The overarching approach in my research is the so-called synthetic methodology: understanding natural phenomena by realizing them in artificial systems and, at the same time, seeking how to turn the artifacts into applications (see Hoffmann & Pfeifer 2018 for more details [pdf]).

Modeling development

We conduct or collaborate on empirical studies with infants and apply embodied computational models (baby humanoid robots) to get insights into the mechanisms of sensorimotor development and the acquisition of "body know-how". Please consult the Modeling infant sensorimotor development tab for details.

To obtain maximum information from dense recording of infants, we are developing various "AI tools for developmental science", including body pose and gaze estimation, motion retargeting, and video deidentification. See the tab AI tools for developmental science. 

Human-robot interaction and artificial touch

We target two application areas where robots with sensitive skins and multimodal awareness provide a key enabling technology:

(i) safe and intelligent physical man-machine interaction and collaborative robotics (Co-bots and physical HRI tab) 

(ii) automatic self-calibration (Artificial touch and self-calibration tab)

Active perception for robot manipulation

Robots can learn about objects by interacting with them: a touch reveals a hidden surface, a squeeze indicates softness, and a tap produces sounds that help identify a material. We develop methods that combine vision, touch, force sensing, and sound to estimate object shape and physical properties. A central question is how a robot should choose its next exploratory action to reduce uncertainty and gather information useful for manipulation. See the Active perception for robot manipulation tab. 

Embodiment and morphological computation

I am also interested in how the body morphology (shape, material properties etc.) can contribute to behavior and cognition in natural and artificial agents. This is typically referred to as embodiment (see e.g., Hoffmann & Pfeifer 2018 [pdf]  and 2011 [pdf]), but sometimes also morphological computation - see the Morphological computation tab. 

Minimally cognitive robotics - from locomotion to cognition

The work that formed the basis for my dissertation (defended in 2012 - pdf) was conducted at the Artificial Intelligence Laboratory, University of Zurich, under the supervision of Prof. Rolf Pfeifer. The overarching theme was to investigate the emergence of minimally cognitive phenomena in different robots that - equipped with a multimodal set of sensors - learned to extract the regularities from their interaction with the environment, stored them and later deployed them to improve their performance. The key concepts were body schema, forward internal models, and sensorimotor contingencies. The main ideas and implications of this research are summarized in a book chapter ([Springer][preprint]). More details can be found on the From locomotion to cognition subpage.