Systems AI: towards autonomous robots

Activities that are intuitive and natural for humans are a big challenge for robots. Even the way in which a robot moves needs to be trained. And without precise programming or extensive training, robots cannot judge how hard to grasp an object without damaging it. This makes it difficult to use them efficiently and safely in dynamic environments, especially in direct contact with people.

How autonomous robots of the future can be efficiently trained in body control and object manipulation is demonstrated by DFKI's research department Systems AI for Robot Learning (SAIROL). Simulations play a central role here, as they serve as a basis of experience for intelligent autonomous robot systems. They can be used to train neural networks and develop model predictive control algorithms that will enable the robot to perform activities autonomously. At the booth, a humanoid robot will be remotely controlled to interact with various objects using an exoskeleton and VR goggles. In parallel, a digital simulation will be generated that can be used as a basis for further training.