Max Planck Research Group for Autonomous Vision
We are interested in computer vision and machine learning with a focus on 3D scene understanding, parsing, reconstruction, material and motion estimation for autonomous intelligent systems such as self-driving cars or household robots. In particular, we investigate how complex prior knowledge can be incorporated into computer vision algorithms for making them robust to variations in our complex 3D world. You can follow us on GoogleScholar (paper email alert), on YouTube (video email alert) and on Facebook. Pictures from recent group activities can be found in our gallery!
Max Planck Research Group for Autonomous Learning
We are interested in autonomous learning, that is how an embodied agent can determine what to learn, how to learn, and how to judge the learning success. In particular, we focus on learning to control a robotic body in a developmental fashion. Artificial intrinsic motivations are a central component that we develop using information theory and dynamical systems theory. We work on reinforcement learning, representation learning, and internal model learning.
We conduct research to understand neurocontrol and mechanical principles of dynamic legged locomotion.
The Dynamic Locomotion Research Group at the Max Planck Institute for Intelligent Systems in Stuttgart is run by Dr. Alexander Spröwitz. Robotic Biomechanics is his area of expertise, as he sees himself as a Roboticist and Biomechanial Engineer.
Animals walk dynamically and efficiently, elegantly and adaptive. Their movement is a carefully orchestrated interaction of muscles and tendons, that nature perfected in the course of evolution. Spröwitz and his team of four PhD students use robots to help them understand animals and their locomotion.
How are powerful, functional linkages created from networks of muscle-tendons? Which leg activation patterns help the animal to robustly and effectively move? The scientists use the blueprints they see in nature and create robot-models that replicate the structure of the animal – with the robot being exposed to the same forces (for instance gravity) and impacts as its living counterpart.
So how do the different parts, the bones, muscles and tendons interact with each other? An advantage the robot has over the animal is that the scientists can observe and test each component individually. That´s not possible with a dog or a cat – soft tissue,
the skin, and the complex 3D architecture of muscle-tendons block the view.
A good example is the spider. It walks and jumps by using a hybrid system of muscles and hydraulics. To be able to move its joints, hydraulics is at play – catapulting it into the air as high as 20 times her own body height. How and where does it build up this much pressure? And how does the joint membrane expand like a hand fan? Spröwitz and his team closely look at the spider, replicate its leg and then make conclusions how animals successfully overcome the limitations of miniaturization and limited muscle power.
Same goes for the Cheetah-Cub robot, weighing 1.4kgs and about the size of an average domestic cat. How can this robot master uneven terrain - like step-downs - just like any cat would do? How much energy must the robot just like the cat apply to be able to walk? To understand its locomotion, Spröwitz and his team built a Cheetah-Cub robot to find out.
Our group has broad interests in the interaction of optical, electric, and magnetic fields with matter at small length scales. We work on new 3-D fabrication methods, self-assembly, actuation, and propulsion. We have observed a number of fundamental effects and are developing new experimental techniques and instruments.
The Independent Max Planck Research Group on Probabilistic Numerics
Numerical Problems --- linear algebra and optimization, integration and the solution of differential equations --- are the computational bottleneck of artificial intelligent systems. Intriguingly, the numerical algorithms used for these tasks are also compact little intelligent agents themselves. They estimate unknown / uncomputable quantities by observing the result of feasible computations. They also actively decide which computations to perform.
The Research Group on Probabilistic Numerics studies this philosophical and mathematical connection between computation and inference. We aim to build a theoretical understanding of numerical computer algorithms as agents acting rationally under uncertainty. We analyse existing algorithms from this viewpoint, and propose novel algorithms that provide functionality for key computational challenges in the science of Intelligent Systems.
Max Planck Fellow Group
We work on the theoretical analysis of machine learning algorithms. Our current focus is on comparison-based learning algorithms and on algorithms on random graphs and networks. The group is lead by Ulrike von Luxburg, the funding comes from a Max Planck Fellowship.
The groups by Ulrike von Luxburg are distributed between the Max Planck Institue and the University of Tübingen, our main webpage is the one at the university .
The Max Planck branch of our group consists of the following people: