Robotics and Control

While machine learning is the methodology used to push robot performance to the limits, proper knowledge of robotics systems is crucial to successfully apply learning techniques to robotics platforms. Our research in robotics focuses on exploiting the structure of mechanical systems to improve the low-level control performance, impose safety constraints, and efficiently solve manipulation tasks. In particular, we believe that the manipulation problems are of great interest, as a wide variety of new sensor technologies, mechanical improvements, and increased computational power will allow the exponential growth in robotic manipulation capabilities. Below one can find further details on the following topics: Safe and Robust Control, Learning to Plan, Tactile Manipulation, Differentiable Physics, Sim2Real Transfer.

Safe and Robust Control

Safety is the crucial property of any real-world system. When deploying hardware in a dynamic environment, we require the autonomous system to prevent damage to people, surroundings, and itself. To achieve this result, we focus on techniques to maintain the safety of autonomous and learning systems by exploiting all knowledge available from the platform. To ensure that the low-level control loops are working under any environmental conditions and robot state (TODO here I mean wear & tear), we develop robust control techniques able to maintain good performance under any circumstances. (TODO maybe a bit more specific from Joe).

Learning to Plan

Planning is crucial in many robotics applications. From low-level kinodynamic planning of rapid motions to long-term and complex assembly tasks, planning is often an irreplaceable component of intelligent systems. However, standard planning techniques are often limited both by the scale of the problem, requiring approximate reasoning instead of an exhaustive search, and by the environment dynamics, requiring reacting fast to unexpected events. In both cases, learning techniques have shown the ability to offer scalable and reactive solutions that are difficult or infeasible to implement using standard planning techniques. The unique mixture of robotic knowledge and machine learning allows us to benefit from both worlds and exploit planning methods for complex and highly reactive tasks.

Tactile Manipulation

Many current robots lack fine manipulation skills. A major reason for this is that most industrial robotic arm-hand systems do not receive sufficient feedback about contact with the object. Neuroscience has shown that such feedback from tactile sensors is a critical component in the human ability to perform such tasks. Therefore, we aim to equip our robots with tactile sensors and use the sensor feedback for robot control. Research issues include how to interpret the signals from the sensors (that are often noisy and high-dimensional) and how to include such signals in a robot control loop. Among others, we aim to address such issues by using supervised, unsupervised, and reinforcement learning techniques. (TODO copied from the old one as it fits, but maybe needs some suggestions from Boris)

TODO: Boris, Tim, Niklas

Sim2Real Transfer

TODO