Inhalt des Dokuments
Update 23/07/2016: Our paper has won the Best Systems Paper award! Congrats to my colleagues and co-authors!
I'll be giving a talk about Learning with Side Information at the Berlin Machine Learning Group Meetup on May 2nd, 7pm at Amazon ML. Unfortunately the event is already fully booked, but there is a waitlist available. (But the main reason for this is probably the free pizza, not my talk ;-)
We have made the paper describing the perception system of our winning entry to the Amazon Picking Challenge 2015 available as a technical report:
Final version of our ICRA16 paper on integrated tracking, segmentation and shape reconstruction of articulated objects is online:
Paper and code for our paper on learning with side information are available on arXiv/github:
I am interested in robots who learn autonomously from their own interaction how to successfully manipulate their environment. Learning happens on many levels: I am particularly interested in the boundary between low-level sensorimotor learning and high-level reasoning. My recent work deals with the autonomous acquisition of high-level relational models for discovery of an object's degrees of freedom  (PDF, 400,5 KB), and methods to learn about the position and shape of articulated objects from vision .
More recently, I got interested in how additional modalities allow to learn more task-specific representations in order to make learning more effective .
Previously I have been working at the labs of Luc Steels at Sony CSL Paris and the VUB in Brussels on evolutionary linguistics (c.f. this introductory paper). I was working on a project dealing with autonomous grounding of action language , and linguistics projects on computational modeling of grammatical agreement  (PDF, 729,0 KB) as well as of complex grammatical features of Polish .
For my master thesis, I was working at the Neurorobotics lab under the guidance of Manfred Hild. I was studying the question how a humanoid robot can acquire a compact representation  of its prioprioceptive sensory input, using Slow Feature Analysis (c.f. Scholarpedia). The main goal was to learn to classify the robot's postures in an unsupervised manner.
Junior researcher at the Robotics & Biology Lab, TU Berlin (Oliver Brock).
In May 2015, our team RBO won the inaugural Amazon Picking Challenge, a grasping challenge with 26 teams from all over the world
Research assistant at the AI Lab, Vrije Universiteit Brussel (VUB) (Luc Steels).
Diploma in computer sciences with minor subject philosophy at the Humboldt-Universität zu Berlin (HU Berlin).
My diploma thesis Applications of Slow Feature Analysis in Humanoid Robotics received the Best Thesis Award 2011 (Institutspreis 2011) by the HU Berlin Department of Computer Science.
Internship at the Sony Computer Science Laboratory, Paris (Luc Steels).
Student assistant at the Labor für Neurorobotik, Humboldt-Universität zu Berlin (Manfred Hild).
Zusatzinformationen / Extras
Tel. +49 30 314-73 110
Fax +49 30 314-21 116
Raum MAR 5.032
Sekretariat MAR 5-1