With our research we connect basic research in Artificial Intelligence – in particular related to knowledge representation and reasoning – to technical systems in order to realize intelligent systems which interact with humans. To this end we often employ hybrid approaches to AI and we are particularly interested to investigate how different AI components can be integrated. 

Research highlights:

Qualitative Spatial and Temporal Reasoning (QSTR)

Aside regular contributions to field we develop the spatial reasoning toolbox SparQ or see our recent survey on existing formalisms. Using QSTR techniques in our BamBird agent we won the AI Birds competition in 2016.

Hybrid AI for Depenable and Safe Applications of AI

Most real-world application involve processing of sub-symbolic data, e.g., obtained from perception. Connecting the sub-symbolic to the symbolic layer (symbolic grounding) is a longstanding challenge in AI research. A misalignment easily introduces significant problems as we easily notice in practical experiments, for example in context of our BamBird agent. With our basic research in symbolic AI we aim to contribute to making AI applications more dependable, for example by applying qualitative spatial reasoning to online verification or by intergrating machine learning and symbolic reasoning.

Spatial Language and Situated Interaction

We are keen to connect spatial knowledge representation and reasoning to language in order to achieve a intuitive interactions with intelligent systems. Currently, we pursue the automatic interpretation of place descriptions in context of a DFG project within the priority program on volunteered geographic information (VGI).