Today, human space exploration faces persistent and new challenges. Most prominently limited crew size, limits on crew time and utilization, as well as the costs of 24/7/365 operations are the main drivers for innovations towards increased autonomy. Furthermore, in the future, when considering destinations beyond lunar orbit, signal delay simply prevents the conventional operational concept of real-time monitoring and control for crewed missions. To overcome those challenges, this paper will introduce METIS, the Mars Exploration Telemetrydriven Information System. METIS is a novel intelligent assistant enabling autonomous operations of crewed spacecraft based on artificial intelligence (AI) and machine learning (ML). METIS is currently being developed at the Columbus ControlCenter (Col-CC), which is part of the German Space Operations Center (GSOC). On behalf of the European Space Agency (ESA), Col-CC is responsible for the operation of the Columbus module of the International Space Station (ISS). Columbus was launched on February 7, 2008, docked to the ISS a few days later on February 11 and has been a part of the ISS since then. It is the basis on which METIS is developed. The assistant is designed after John L. Boyds OODA (Observe-Orient-Decide-Act) loop for human decision making, where each function in the model is developed as an individual computerized agent. This paper will introduce the high-level system architecture of this multi-agentsystem (MAS), as well as human interaction embedded in the system. Furthermore, we will show each of the agents in greater detail, their functions and the interaction between the agents. Since some agents rely on machine learning (ML) from the field of artificial intelligence (AI) we use multiple data sources from the Columbus module, including but not limited to on-board telemetry, operational data files (ODFs, i.e. procedures), and anomaly reports, to train and test the system. We will highlight how these data sources are brought together and the database we deploy to enable access to the data in a machine-readable format. Since METIS is intended to be used by both ground personnel, as well as on-board crew, we will show how METIS is planned to be utilized within Col-CC and on-board. For this we envision a step-wise approach, where METIS is first used and tested in conjunction with the Columbus simulator to perform certain ground and on-board activities and react to a set of known anomalies. In this first step ground controllers will still be heavily involved in development, to identify areas that need further improvements and further increase the capabilities of the system. This work represents a significant step towards full crew autonomy by supporting quick and reliable decision making in the highly dynamic environment that is human space exploration. AI/ML enabled solutions to new and old challenges have the potential to revolutionize the space sector and METIS represents one possible contribution to make space more accessible to humans, increase system reliability, reduce cost of operations and help prepare for deep space exploration beyond lunar orbit.
