Towards Semantic Grasp Planning using Ontology-based Geometry and Capability Models

abstract

Robotic manipulation in small and medium-sized enterprises (SMEs) is challenged by high product variability, small lot sizes, and frequent task reconfigurations. Under these conditions, traditional data-driven grasp planning methods are often infeasible due to their reliance on large, annotated datasets and stable object libraries. This paper presents an architecture for ontology-based semantic grasp planning that enables efficient and intuitive manual grasp pose specification without requiring extensive expertise in geometric programming or machine learning. By leveraging semantic representations of object geometries, gripper capabilities, and task-specific constraints, the system supports context-aware grasp candidate generation and prioritization. While the interpretation of the semantic models identifies suitable grasp types and involved geometric faces of objects and grippers, an optimization step generates suitable 6D poses that meet given constraints. As a result, the complexity of manually instructing robot systems for manipulation tasks is reduced and can be efficiently carried out by non-expert users.