Robontics 2025 - Ontology-Guided Multi-Modal Perception for Trusted and Explainable Robotic Action

This repository contains the implementation resources, ontological models, and experimental data accompanying the paper “Ontology-Guided Multi-Modal Perception for Trusted and Explainable Robotic Action. The dataset is based on the GOOSE multi-modal navigation benchmark, which includes synchronized RGB, depth, and telemetry sequences recorded for autonomous navigation and driving research. The provided files include preprocessed perception masks, ontology assertions (RDF/OWL-Lite), SHACL validation rules, and reasoning logs used in the experiments. The code implements the complete neuro-symbolic reasoning pipeline: perception (SegFormer/DeepLabV3), ontology mapping, SHACL-SPARQL constraint validation, and explanation generation modules.All experiments were conducted in a reproducible Colab environment with open-source dependencies (PyTorch, Transformers, rdflib, pySHACL). The resources are released under a permissive license for academic and research use, ensuring transparency, traceability, and reproducibility.