Robot Situation and Task Awareness Using Large Language Models and Ontologies

摘要

Robot situation and task awareness requires a deep understanding of the environment, the domain knowledge, and task planning. We present a novel framework that integrates ontologies, Large Language Models (LLMs), and the Planning Domain Definition Language (PDDL) to enhance the comprehension capabilities of robotic systems. The framework employs an LLM to extract structured knowledge from natural language descriptions provided by a human user, populating an OWL ontology that captures relevant objects, properties, and relations. This populated ontology is then used to parse a PDDL Domain file and generate a corresponding PDDL Problem file to solve particular planning problems. This research contributes to the intersection of knowledge representation, natural language processing, and automated planning, providing a solution for intuitive human-robot interaction through LLMs.