Autonomous robots are currently active across various fields, executing complex tasks. While their use in dynamic environments holds limitless potential, unresolved challenges in reliability and trust still pose significant risks.
This article, which correspondence author is Esther Aguado (UPM), offers a summary of prominent and recent studies that employ knowledge-based methods to enhance robot autonomy. One of the main goals is to conduct a systematic analysis of recent and relevant projects that use ontologies for autonomous robots.
Some highlights from this article:
- We consider ontologies to be a vital component in enhancing robot autonomy. However, challenges remain, including the limited spread and alignment of various ontological methods, highlighting significant opportunities for further research in this area.
- We believe that future engineering-grade, knowledge-driven autonomous robot software platforms should also be capable of providing three characteristics: explainability, reusability, and scalability.
- Despite most of the work being focused on categorization, decision making, and planning, monitoring and action coordinaton are critical processes with respect to robot control, especially for robust and reliable operation.
- We consider explicit knowledge to be essential for supporting autonomous robot operations in unstructured environments. Although challenges remain in terms of reliability, safety, and explainability to meet the expectations of researchers and industry, the progress made by various projects in enhancing autonomy through ontologies demonstrates the promise of this approach.
Part of this research relates to the analysis of ontologies present in Coresense deliverable D1.2. Analysis of Ontologies and of Ontological Formalisms. Although many of the ontologies reviewed in this document are not sufficient on their own to support CORESENSE, they all form important contributions to our ontology effort. No ontology can cover all human knowledge, and as large and rigorous as SUMO (Suggested Upper Merged Ontology) is, it will still require extension. To that end, have reviewed existing ontologies in the autonomous system domain to create our Coresense ontology, extending its conceptual inventory when any concept found in other ontologies is found to be missing or insufficiently defined in SUMO.
In conclusion, we propose extending existing ontologies formalisms expressed in highly expressive logical and systematic languages as a strategic approach. This choice enables shared, explicit, and formal definitions of concepts to support deep understanding and awareness during run- time execution. The Coresense ontology models will encapsulate the essential knowledge possessed by the agent, facilitating the integration and composition of diverse information types, such as DAE, kinematics, dynamics, and data-driven models. It will serve as a foundation for deriving additional insights and generating appropriate behaviours, reasoning about which representations are best suited for specific situations.
You can access to the article here.
