Combining SMT with ML
for AI Planning in Real-World Applications 
 

23 October 2025 | Aula Stringa - Online | 11:00 | René Heesch  (Helmut Schmidt University) 

Abstract
Planning as satisfiability has proven to be a powerful and efficient approach for solving complex planning problems, ranging from classical to numerical domains. However, despite its algorithmic strength, a critical bottleneck persists: the need for pre-defined, manually specified domain models. This assumption significantly limits applicability in real-world settings, where such models are rarely available or are prohibitively costly to construct. We address this challenge by integrating planning as satisfiability with machine learning techniques to automate the acquisition of domain models directly from data. Our approach extends beyond learning symbolic representations, such as action models, to directly integrating trained ML models into the domain description, for example, instead of a symbolic description of the effects of actions. 

Bio

René Heesch is a Senior Scientist at the Chair of Computer Science in Mechanical Engineering at Helmut Schmidt University, Hamburg, Germany. He leads the chair’s research group, which conducts projects in collaboration with the German Federal Armed Forces. His PhD research focuses on integrating Satisfiability Modulo Theories (SMT) with Machine Learning to advance AI planning for real-world applications.