The main challenge is that autonomous systems in SFGs have to work with limited and highly diverse data. While traditional machine learning methods require large and consistent data sets, these networked production environments continuously generate new, heterogeneous, and sometimes incomplete data streams. 

The systems must therefore learn to extract relevant information from multimodal sources such as sensor, machine, and production data and integrate it into a consistent model. At the same time, they must be able to retain existing knowledge, learn new skills, and selectively forget superfluous information. 

A particular challenge here is the ability to learn through observation, for example when robots take on new tasks by imitating human actions. Equally important is the development of robust continuous learning methods that do not lose performance even in the event of structural changes or interruptions in the manufacturing environment, but can readjust themselves in a targeted manner or restart the learning process if necessary.

The research approach combines the development of efficient learning methods with self-optimization strategies. On the one hand, the aim is to create models that can extract reliable insights from a small amount of multimodal data and derive generalizable structures from them.

On the other hand, the systems should be able to develop independently, learn new skills, and improve their performance during operation. Through the use of imitation learning, autonomous units can quickly adopt new behaviors, while continual learning methods ensure that once acquired knowledge is not lost but is specifically expanded.

In the long term, this research will be supplemented by aspects such as explainability, uncertainty, and risk assessment in order to make decision-making processes more transparent and reliable. In close cooperation with the areas of adaptive control, system health, network security, and additive manufacturing, this will result in an integrative approach that forms the basis for intelligent, resilient, and self-optimizing production systems.t