How can AI systems better apply their learned behavior to new and previously unknown situations? This is a question to be addressed by Dr. André Biedenkapp in his future research at the Karlsruhe Institute of Technology (KIT). Biedenkapp has been awarded funding for an Emmy Noether Group from the German Research Foundation (DFG) for his research on improved generalizability in reinforcement learning methods and will receive EUR 1.2 million over the next three years. The DFG has committed to further funding of EUR 920,000 for three more years subject to a successful interim evaluation after the initial period.
"The Emmy Noether award helps outstanding young researchers to become academically independent at an early stage in their careers," said Professor Stefan Hinz, Vice Provost Early Career Researchers at KIT. "His work on developing adaptive AI systems makes Dr. André Biedenkapp a prime example of an excellent young researcher at KIT."
Reinforcement learning (RL) is an artificial intelligence (AI) learning paradigm in which an AI agent learns how to behave in a specific environment by trial and error. Feedback in the form of rewards helps a system to repeat desired behaviors and avoid inappropriate ones. This is an especially powerful method for problems in which decisions must be made sequentially, e.g. in robotics, logistics, or resource management.
However, a key problem with traditional RL approaches is that the learned strategies often depend strongly on the training environment. Even small changes can mean that an AI agent no longer knows how to behave appropriately. "Today's RL agents work superbly under the conditions they've been trained for, but they reach their limits quickly when those conditions change," said Biedenkapp, who is working at the University of Freiburg through August 2026. From September 2026, he will be leading the newly funded DFG Emmy Noether Group "From Mediocre to Masterful Generalists: The Power of Context in RL" at KIT's Institute for Anthropomatics and Robotics.
More Context for More Robust Learning Processes
The Emmy Noether Group's goal is to extend RL training methods so that AIs become more robust and adaptable. To do so, Biedenkapp's team will make use of additional information about the environment or world in which an agent acts. In this way, an AI can learn which behavior is best suited to which situation and then apply that knowledge to similar unknown situations later.
In the long term, this approach could be a key step toward increased use of RL in real-world applications. Many RL-based AI systems have thus far had to rely on very exact simulations of real environments, but the required simulators are complicated, expensive, and difficult to implement for complex scenarios. "If RL-based systems could generalize better, it would no longer be so important to simulate every possible situation perfectly. That would expand the range of possible applications for this technology considerably," Biedenkapp said.
About the Emmy Noether Program
The Emmy Noether Program gives exceptionally well qualified scientists in the early phase of their career the opportunity to qualify for a university professorship by independently leading a research group over a period of six years.
