Can I do it

TL;DR

The paper tackles the challenge of enabling a robot to answer whether it can perform a given task under current conditions by using an online, ontology-driven knowledge base that encodes engineering knowledge about configurations, components, and performance. It combines a hypergraph-based ontology with explicit realizing and processing relations to infer viable configurations per behavior and to predict task-related performance from live sensor readings. The Self-X framework contributes a practical pipeline for modeling components, configurations, and behaviors, and demonstrates a real-life SPOT robot implementation that selects search strategies based on predicted performance. The work advances runtime self-assessment and self-reconfiguration by bridging engineering knowledge, environmental context, and empirical component performance, paving the way for more adaptive and reliable autonomous systems.

Abstract

Knowledge about how well a robot can perform a specific task is currently present only in engineering reports which are inaccessible to the robot. Artificial Intelligence techniques, such as hypergraphs and automated reasoning, can provide such engineering knowledge online while enabling updates in the knowledge with new experiences. This requires a sound knowledge structure and maintenance routines for keeping this knowledge-base about the robot's capabilities truthful. A robot with such up-to-date information can reason about if and how well it can accomplish a task. This article introduces a knowledge representation that combines an ontology on system engineering, a deductive reasoning on the connections between system components, and an inductive reasoning on the performance of these components in the current system configuration. This representation is further used to derive the expected performance for the overall system based on a continuous evaluation of the actual performance per component. Our real-life implementation shows a robot that can answer questions on whether it can do a specific task with the desired performance.

Figures (25)

• Figure 1: Two alternative system configurations to search for persons.
• Figure 2: Setup of an autonomous robot that is planning how to conduct a task as a series of different behaviors.
• Figure 3: An illustrative example of a hypergraph, in its original representation (a), and in a plain graph representation (b).
• Figure 4: The highest level concepts in the model for a Component producing a Creation while requiring other Creations to call for. In case a box has a dashed line it means that the concept is already present in the model and thus that, in the actual ontology and implementation, its edges should be connected to the original concepts having the solid line. However, when illustrating the ontology for a system engineering the above view is more clear compared to an illustration of the true ontology without such duplicates, as it illustrates a Creation at the 'input' and a Creation at the 'output' of the Component which is more in line with the domain of system engineering. Also note that a has-link is depicted as a closed arrow.
• Figure 5: The hierarchy of sub-classes for Creation, Component, Require and Property. Every open-ended arrow is a sub-relation, while a dashed arrow implies that additional types can be included. Entities are illustrated as a rectangular box, relations as a diamond and attributes as an ellipse.