Space
Google and NASA Develop Autonomous Robotic Medical Assistant for Deep‑Space Missions
Google and NASA are working together to create an autonomous robotic medical assistant that will help astronauts handle health emergencies and perform routine tasks on future deep‑space missions. The partnership comes as NASA’s Artemis program moves toward more challenging lunar missions and prepares for the first crewed Mars flights. The robot is intended to address the challenge of providing advanced healthcare when real‑time communication with Earth is impossible. When not performing medical duties, the system is designed to take on routine operational tasks such as hardware maintenance, equipment transport, and inventory management.
The effort brings together specialists from Google Cloud, NASA’s Human Research Program, the Johnson Space Center robotics team, and the Jet Propulsion Laboratory. It builds on the Crew Medical Officer Digital Assistant, an AI platform that Google and NASA began evaluating in 2025 to enable crews to diagnose and manage medical issues independently when signal delays prevent rapid consultations with flight surgeons.
Jim Kelly, vice president of Google’s public sector division, said an astronaut on a future Mars mission could face round‑trip communication latencies of up to 44 minutes, making real‑time medical guidance nearly impossible. The project is currently experimental; laboratory testing focuses on validating foundational robotic manipulation and general utility tasks. Fully autonomous medical intervention in microgravity remains a long‑term goal.
The robot is envisioned as a versatile payload handler that can manage cargo, logistics, and equipment. When not needed for routine maintenance, it can pivot to medical support—for example, retrieving and preparing an ultrasound machine if an astronaut experiences severe back pain.
To operate safely in confined spacecraft, the system uses vision‑language‑action models and spatial awareness tools adapted from satellite situational awareness intelligence. Testing at Johnson Space Center’s iMETRO facility used Google DeepMind’s MuJoCo simulator and an eye‑to‑hand camera configuration, achieving an initial 85% success rate in a pick‑and‑place task.
Processing complex tasks requires significant computing power. The team trained a 4‑billion‑parameter Pi0 model on Google Cloud G4 instances in 2.5 hours and demonstrated that the fine‑tuned model can run locally on compact, energy‑efficient edge devices. Over the past two years, the team tested about 15 different edge devices to identify those suitable for lightweight, low‑power operation required for future Mars missions.
The ultimate goal is a robotic system that can execute a variety of tasks via simple voice instructions under direct astronaut supervision, providing physical assistance without replacing human decision‑making. This collaboration aims to support increasingly difficult missions and advance scientific discovery in space.