Together with academic partners and universities, we research cutting-edge machine learning methods, robotics simulations, and explainable AI workflows. Not all research outcomes are integrated into the current TRYB platform yet – here we provide insights into our funded research initiatives.
In cooperation with Chemnitz University of Technology, a simulation-based training platform for intelligent manufacturing robots is being developed. Unlike training in reality, this virtual pre-training can be accelerated many times over.
The results of the virtual training can be transferred to the real robot, which then only needs to be retrained in a real environment for a fraction of the time currently required.
This project explores an innovative system that enables non-experts to create complex robot programs without prior programming knowledge using advanced language models.
Language models have the potential to generate robot programs directly from natural-language instructions. In addition, reinforcement learning is used on a digital twin of the robot cell to train complex movement sequences that cannot be generated by the language model.
The aim is to overcome barriers to using robotics and to increase production flexibility, especially for individual customer requirements. The project will produce a demonstrator that forms the basis for a commercially viable software product.
Around facilitating the development of AI systems in Bavarian SMEs, this project explores an AI tool in the form of interactive and controlled automation of the entire machine-learning lifecycle.
The focus is on two industrially relevant application types, with image and time-series data posing particular challenges in terms of data preprocessing and training. By integrating methods of explainable, collaborative, semi-automated machine learning (AutoML) and quality controls, the project aims to increase transparency and trust in automated ML processes.
The main elements of the project are extensive requirements analysis, development of a demonstrator, and user studies to validate and optimize the effectiveness, usability, and applicability of the developed methods.
