FZI Forschungszentrum Informatik

FZI Forschungszentrum Informatik

The FZI Research Center for Information Technology is a non-profit institution for applied research in information technology and technology transfer. For the mobility of the future, scientists at the FZI are developing automated cooperative and interacting mobility systems – from autonomous vehicles to innovative concepts for public transport and AI-supported procedures for the management of personal mobility. Cooperating closely with companies from the mobility sector, the FZI bridges the gap between basic research and practical application. The current focus is on novel mobility concepts and applications with particular emphasis on public transport. While taking legal aspects into account, FZI scientists develop from the initial idea to system design, algorithms and testing innovative mobility solutions for the transport of people and goods. Furthermore, they work on the design and construction of supporting infrastructure systems. The security and robustness of the developed solutions is ensured by virtual, semi-virtual and real testing.

In the FZI House of Living Labs the FZI operates several vehicles with clearance for fully automated driving in road traffic according to §70 StVZO and has access to real laboratories such as the Test Area for Autonomous Driving Baden-Württemberg.

The extensive experience and competence in AI systems and methods of machine learning and probabilistic description is passed on to numerous joint projects as well as in diverse direct contracts with the industry. Main focus areas are robust, AI-based perception and decision making, a continuous probabilistic situation interpretation as well as maneuver and trajectory planning for autonomous vehicles, taking into account machine learning methods as well as their safe application and transferability.

The overall goal is to make mobility safe, sustainable and comfortable. Within the project KI Delta Learning the FZI focuses on the application of active and continuous learning methods to maneuver and trajectory planning and the applicability to reinforcement learning techniques. Furthermore, AI-based domain adaptation from synthetic to real sensor data is investigated.