An evidence-driven human-agent-robot teaming architecture for Earth-independent anomaly triage, demonstrated end to end with Reachy Mini and MARS
The work presents an evidence-driven human-agent-robot teaming architecture for Earth-independent anomaly triage in deep-space crews: agentic AI acts only as a coordinator over explicit state and bounded, inspectable services, with a triage state manager holding hypotheses, evidence provenance and uncertainty, a crew-facing embodied agent eliciting observations and explaining assessment changes, and a mobile robot acquiring localized evidence; two scenarios, a Mars-transit case anchored on an actual ISS ammonia false alarm and a lunar-base power-interface anomaly, illustrate the architecture, and a hardware-in-the-loop prototype using Reachy Mini and an Innate MARS robot exercises the closed evidence loop end to end.