New control strategies for industrial robots
DFG project
In the project “Reinforcement learning for efficient path planning of parallel robots when handling flexible objects”, Prof. Tobias Glasmachers, Chair of Theory of Machine Learning at the Faculty of Computer Science, and Prof. Bernd Kuhlenkötter, Chair of Production Systems at the Faculty of Mechanical Engineering, are researching efficient control strategies for industrial robots. The project is funded by the DFG for three years.
Due to their lightweight design, parallel robots are an ideal tool for sorting tasks (pick-and-place). They are used, for example, in the food industry for packaging products.
This type of robot is characterized by a closed kinematic chain in which several arms control a platform with a purpose-specific tool. This allows it to work particularly fast and precise. The dynamic properties of parallel robots ensure high speeds and accelerations on the one hand, but pose a huge challenge for the design of optimal control strategies on the other hand.
This is because in parallel robotics, the user typically defines the movement path by chaining together linear paths in Cartesian coordinates. This results in a restricted, non-optimal robot path. The aim of the project is to achieve the most efficient and optimized path planning possible. To achieve this, Glasmachers and Kuhlenkötter use machine learning methods from the field of reinforcement learning. In particular, the researchers see great potential for efficiency gains by adapting the path to the position and orientation of the objects to be transported.
System vibrations also play a major role here, as parallel robots are very fast and lightweight compared to other types of industrial robots, and they are often used for transporting flexible objects. Complex simulations are used in the project to handle these high dynamics, which are validated on the real hardware using state-of-the-art sensor technology.