Speaker
Description
Satellite launch traffic has rapidly increased following the advent of satellite mega-constellations, and the growing population of active satellites necessitates measures to ensure that space remains safe for long-term use. While major space organisations have independently implemented policies for space sustainability, it may yet be necessary to develop governmental policies for satellite and debris management. Approaches for debris mitigation can be modelled via simulation, establishing the effects of policy-led restrictions, but interpreting their outputs often requires specialist knowledge, posing difficulties for policy-makers' understanding.
To make space environment forecasts more accessible, we propose a large language model (LLM)-based “digital expert” and provide an example for the NEtwork model for Space SustainabilitY (NESSY). NESSY simulates interactions between satellites and debris under user-selected policies and returns mathematically-grounded “health indicators” which require expertise to interpret. We use the Llama LLM family to provide these interpretations via a chat platform. As this requires technical knowledge that “base” Llama lacks, we use retrieval augmented generation (RAG) to draw information from curated documentation. This allows us to leverage the flexibility of LLMs to tailor explanations to the user’s expertise level while requiring support by factual information.
We assess our model using twenty evaluation questions, scored on a five-point scale; “base” Llama averaged 1.35 points over all questions, while our method averaged 4.60 points, demonstrating clear potential for “out-of-the-box” LLMs to be adapted as digital experts. We are presently developing similar models for other simulations to provide integrated tools that allow informed assessment of policy outcomes.
