Toward Human-Aware World Models for Astronaut-AI Teaming

Future space missions increasingly envision robotic systems operating as collaborative teammates alongside astronauts. Existing human-AI teaming systems typically rely on fragmented representations and reactive adaptation and rarely maintain persistent predictive models of the coupled human-AI-environment system. World models provide a principled foundation for this capability through learned latent dynamics; however, current formulations primarily model physical and environmental processes, with human internal variables rarely represented explicitly as part of the predictive state. Consequently, the socio-physical interdependence inherent to teaming remains underrepresented. In this position paper, we propose human-aware world models, a unified latent dynamical framework that jointly represents and predicts the co-evolution of the physical environment, human internal state, AI self-state, and their interaction, and outline architectural principles for their integration into space teaming.