Hybrid AI Systems Grounded on Qualitative Spatio-Temporal Reasoning

Michael Sioutis, Zhiguo Long

Abstract: Qualitative Spatial and Temporal Reasoning, or QSTR for short, is a major research area in AI that deals with the fundamental cognitive concepts of space and time in an abstract, human-like manner. For instance, in natural language one uses expressions such as “Region X is located inside or north of Region Y” or “Task A is scheduled after or during Task B” to spatially or temporally relate one object with another object or oneself, without resorting to providing quantitative information about these entities. In brief, QSTR simplifies complex mathematical theories that revolve around spatial and temporal entities to manageable qualitative constraint languages (calculi), which can in turn give rise to interpretable spatio-temporal representations. Thus, QSTR forms a concise and explainable paradigm for dealing with entities pertaining to space and time, with the potential to boost research in a plethora of domains that can range anywhere from theoretical computer science and logic to practical algorithms and applications. In this tutorial, we take a twofold approach to introducing our audience to the rich research area of Qualitative Spatial and Temporal Reasoning. First, we present the scientific background in detail, mentioning some terminology, key definitions, and problems associated with the field, and follow up with a presentation of the state-of-the-art frameworks that exist for handling QSTR data, focusing on native methods and Boolean satisfiability (SAT) and Answer Set Programming (ASP) approaches. Secondly, and most importantly, we address the gap that exists between QSTR—a symbolic paradigm—and Machine Learning, and bring forward some successful examples of neuro-symbolic integration in the context of spatio-temporal information from the recent literature; we argue for further pursuing this promising research direction and explain the current challenges that need to be overcome for obtaining hybrid AI systems that can be applied to highly active areas such as planning, data mining, and robotic applications.