Despite advancements in machine learning and robotics techniques, there still exists a wide range of activities that robots are ill-equipped to perform. This is known as the tail end of edge cases - novel scenarios that The goal of this research project is to identify, design and integrate ontologies for robotic spatial reasoning for use in these unaddressed scenarios. For example, deducing the correct key from the shape of a keyhole, and or determining one's relative size to a hole. The challenge lies in striking a balance between expressivity (the capacity for the ontology to support perception and reasoning) and complexity (a large set of arbitrary axioms can be dificult to verify and maintain over a period of time). We combat this by using a model-based approach, and designing a curated set of competency questions of increasing dificulty that should build upon a shared ontology signature. These ontologies will then be compiled and integrated into a functioning robotic system as a demo. In addition to being a data-eficient approach capable of solving problems that would be intractable via traditional methods, it is more easily regulated due to its explainable nature (there exists a logical train of thought between the ontology axioms and the final proof result).
CEUR ceur-ws.org
For roboticists, we provide a complementary method to machine learning and mechanics, for robotic spatial reasoning. According to our hypothesis, our method would require significantly less data while still being able to generalize to novel scenarios e.g. reasoning about complementary keyhole shapes, or navigating unfamiliar terrain.
For ontologists, a focus on robotics applications can serve as a new guide for practical, expressive, ifrst-order ontologies. In the past 20 years, we have seen widespread adoption of ontologies as a tool for taxonomy and resolving data conflicts. However, we see the potential for ontologies to play a bigger role in a reasoning capacity. Robotics is a good entry point due to its plethora of existing unresolved problems, and relative ease of access to prototyping.
FOIS 2023 Early Career Symposium (ECS), held at FOIS 2023, co-located with 9th Joint Ontology Workshops (JOWO CEUR Workshop Proceedings
© 2023 Copyright for this paper by its authors. Use permitted under Creative Commons License Attribution 4.0 International (CC BY 4.0). 3. Objective(s) 1. What ontologies are needed for robots to replicate common-sense spatial reasoning? 2. What are the minimum axioms needed to represent a physical space?
We use competency questions to verify and validate our ontological choices,
2. Preliminary example of robotic path finding problem modelled as a Process Specification