My primary research interest is in machine learning (ML) for robotics. More concretely, I am interested in studying how to enable robots to explore environments efficiently and autonomously. Similar to how infants and young children interact with the environment during free play to build intuitive theories about the physical world, we can equip robots with an intrinsic motivation to conduct experiments that challenge their current beliefs about the world. For instance, by using a mental model of the environment, a robot can actively steer its exploration toward observations that maximize information gain and allow it to update its internal model to make better future predictions. My research focuses primarily on model-based and model-free reinforcement learning (RL). In the past, I worked with legged robots (Solo8, Solo12, Bolt Humanoid, Poppy, Boston Dynamics' Spot) and manipulators (Trifinger Platform).
Currently, I also work on the Polybot project as a postdoctoral researcher. The project aims to build a small and flexible robot for sustainable agriculture.
I obtained a master's degree in physics from the Goethe University in Frankfurt am Main, working on a plausible biological model of visual attention in the early stages of the visual cortex at the Frankfurt Institute for advanced studies (FIAS). Later, I obtained a master's degree in computer science, working on zero-shot construction of new object categories in convolution neural networks for image classification. This work was also done at FIAS. Afterward, I went to Tübingen to the Max-Plack Institute for Intelligent Systems (MPI-IS), where I did my Ph.D. in computer science, working on RL for robotics. Currently, I'm a postdoc at the MPI-IS in the Autonomous Learning group and as part of the Polybot project.
Sebastian Blaes
Postdoctoral Researcher in the Autonomous Learning Group at the Max Planck Institute for Intelligent Systems, Tuebingen.
Cycling, running, trekking, mountaineering enthusiast.