Question answering over knowledge graphs for explainable satellite scheduling

Powell, Cheyenne and Riccardi, Annalisa (2023) Question answering over knowledge graphs for explainable satellite scheduling. In: International Astronautical Congress, 2023-10-02 - 2023-10-06, Heydar Aliyev Center.

[thumbnail of Powell-Riccardi-IAC-2023-Question-answering-over-knowledge-graphs-for-explainable-satellite-scheduling]
Preview
 Text. Filename: Powell-Riccardi-IAC-2023-Question-answering-over-knowledge-graphs-for-explainable-satellite-scheduling.pdf
Final Published Version
License: Strathprints license 1.0
Download (537kB)| Preview

Abstract

Schedules for satellitemissions consist of thousands, if not millions, of interconnected activities executing many times across days,months, and years to fulfill mission objectives. The complexity of a schedule can make it difficult for Ground Station Operators (GSO) to understand the relationship between activities as part of a complete mission, especially when schedules have been created by means of an autonomous decision making algorithm. Text-based explanations are helpful in establishing the reasoning behind decisions suggested by algorithms and their impact on the overall execution plan. A Knowledge Graph (KG) can provide the underlying data structure to record what has happened and what is scheduled, as well as the interconnected elements that are impacted by the scheduled activities. The relationship between satellite components, environmental conditions, operational constraints, and mission objectives is complex and highly dimensional, which is not easy for a single operator to manage concurrently. A system that can gather information from a KG, and infer the information stored within, can assist human operators in building a deeper understanding of the relationships of automatically scheduled decisions. A natural language query interface to the KG is the simplest way for a human to interface and extract knowledge. Additionally,manual access to the KG can be provided alongside textual answers, enabling exploration of schedule branches to understand what else can change throughout the mission’s execution. This improves the robustness of a system’s responses to queries and allows for greater flexibility. An overview is therefore examined of how KG and Natural Language Processing (NLP) technologies can be used to facilitate eXplainable Artificial Intelligence (XAI) in satellite scheduling. Namely, how to model the schedule and environment information to be stored in the graph and how to reason over such information by interpreting user queries in natural language. An example is presented demonstrating the capabilities of flexible query interpretation on an Earth Observation (EO) satellite scheduling problem. Finally, the capabilities of KG and NLP technologies to provide more explainable insights into satellite scheduling tasks are discussed in the frame of future possible developments.

CORE (COnnecting REpositories)