Space Agent Project

Overview

Space Agent is one of the core technologies Space Seed Holdings develops in-house. For space-utilisation research, we are building an autonomous experiment-automation platform powered by AI and large language models (LLMs).

When you set out to capture the experimental conditions that only space can offer — microgravity, high vacuum and radiation, all hard to reproduce on the ground — the tightest constraints are the scarcity of launch opportunities and the small number of attempts. Space Agent uses AI to support each stage of research — experimental design, recording, analysis and proposing the next set of conditions — aiming to raise the speed and reproducibility of R&D within those limited opportunities.

What we are working towards

• Support for experimental design — AI assists the step of narrowing down “the next condition worth trying” from prior knowledge and materials data. In space experiments, where the number of attempts is inherently limited, the quality of each design weighs heavily on the outcome.
• Joined-up recording and analysis — embedding AI across recording, data organisation and analysis so that researchers can focus on judgement. Standardised records form the basis for reproducibility and for accumulating know-how.
• Linking orbital and ground-analogue work — we envisage connecting orbital research via ultra-small space modules with preliminary verification in ground-analogue environments, under a shared framework for design and recording.
• Connecting to space IP — we design the experiment platform and the IP strategy as a continuous whole, with an eye to acquiring and using the intellectual property generated by the research.

Technical grounding

Space-utilisation research allows far fewer attempts than research on the ground. To meet this constraint head-on, Space Agent is designed to combine two layers.

• Ultra-small space modules — experiment units for cell culture, materials synthesis and similar tasks are kept compact and standardised so they work within tight limits on power, volume and communications bandwidth. Raising the experimental density per unit payload is what drives the cost-effectiveness of the data obtained.
• Pairing with ground-analogue research — simulated microgravity (clinostats, random-positioning machines) and vacuum/thermal-cycling tests serve as the place to test hypotheses and narrow conditions before committing to an orbital run. Settling the design on the ground first improves the yield of precious flight opportunities.
• Data-driven experimental design — to make the most of each opportunity, we combine a data-driven mindset that narrows candidates from prior data with AI-based support.

We also have manufacturing in space and on the Moon in view. Sintering — consolidating powder into a dense solid in a short time (SPS: Spark Plasma Sintering) — is well-suited to forming materials in space, and we pursue new-materials development combined with materials informatics separately (see the “Next-Generation SPS × MI Materials Project”).

Recent focus

We are exploring how to introduce large language models (LLMs) into the research workflow. Embedding AI across experimental design, recording and analysis, we are developing Space Agent as a platform that aims to combine speed with reproducibility. Autonomous experiment automation is at the concept and development stage, and we intend to verify it step by step in the field of space-utilisation research.