|
Sebastian Bitzer
Sebastian is a research associate working towards his PhD under supervision of Sethu Vijayakumar. He joined the SLMC group in October 2006 and is involved in the EU FP6 project SENSOPAC. Before coming to Edinburgh he studied for an MSc in Intelligent System at University College London where he did a project on spike sorting with Maneesh Sahani at the Gatsby Computational Neuroscience Unit. But he first seriously got into robotics during an internship at the Institute of Robotics and Mechatronics of the German Aerospace Center. There he worked with Patrick van der Smagt on EMG-control of a robotic hand. Sebastian's research interests cover learning with latent variable models, nonlinear dimensionality reduction, sequential probabilistic models, man-machine interfaces, online learning and in general "was die Welt im Innersten zusammenhält". |
|
Dimensionality Reduction for Motion Synthesis and Control: Movement systems with a large number of degrees of freedom, like the human body or a humanoid robot, enable the execution of a large variety of movements. On the one hand this leads to great flexibility and a large repertoire of achievable tasks. On the other hand, the same variability in the system makes generating useful and naturally looking movements a challenging endeavour. We investigate dimensionality reduction as a way to incorporate implicit information given in samples of related movements into the process of movement generation. The aim is to simplify movement generation while still maintaining a level of accuracy required by robotic applications. |
|
SENSOPAC: The goal of this project is to understand tactile cognition by integrating machine learning, haptic robotics, experimental and computational neuroscience, and human psychophysics. At SLMC, we will focus on the machine learning part, which covers the following three problems: how to discover the structure of sensorimotor interactions, how to learn this structure, and how to generalise over multiple contexts. Of particular interest will be how to discover the relevant latent variables that are intrinsic to an object explored manually. |