NASA and Google test AI-powered robot for space missions

NASA and Google Public Sector are developing an AI-powered robotic system designed to support astronauts during future missions to the Moon and Mars. The experimental technology combines autonomous navigation, artificial intelligence and compact computing hardware to assist with routine tasks and potentially medical procedures. The collaboration builds on earlier research into AI-based clinical decision support for astronauts operating far from Earth, where communication delays make immediate medical assistance difficult.

As NASA prepares for increasingly complex missions under its Artemis programme, ensuring astronaut safety while reducing dependence on Earth-based support is becoming a priority. Future missions to Mars could experience round-trip communication delays of up to 44 minutes, making real-time medical consultations with specialists on Earth impractical.

Google Public Sector previously developed the Crew Medical Officer Digital Assistant (CMO-DA), an experimental clinical decision support system intended to provide astronauts with medical guidance in isolated environments. The latest collaboration extends this research into physical robotics.

Robotic assistance for astronauts

The project involves NASA's Human Research Program, the Johnson Space Center Robotics Team, the Jet Propulsion Laboratory (JPL) and Google Public Sector. The robotic system is designed to perform various logistical tasks aboard spacecraft and potentially at future lunar facilities. These include moving equipment, organising tools and handling supplies.

By automating routine activities, researchers hope to reduce astronauts' workload and allow crews to devote more time to scientific activities. The system could also support medical procedures. For example, if an astronaut experiences severe back pain that might indicate kidney stones, the robot could retrieve and prepare ultrasound equipment.

A human crew medical officer would remain responsible for interpreting the images, assessing the patient's condition and making treatment decisions. The researchers emphasise that the technology is intended to provide physical assistance rather than replace medical professionals or autonomous clinical decision-making.

AI models control physical movements

Developing robots capable of operating safely in spacecraft presents specific challenges. Unlike industrial robots working in controlled environments, space robots must navigate confined spaces shared with astronauts and adapt to changing surroundings.

JPL is therefore adapting technologies originally developed for monitoring objects in orbit to help the system understand its three-dimensional environment. Researchers also used Google DeepMind's MuJoCo physics simulator to train and evaluate robotic movements virtually before testing them on physical equipment. During an initial proof-of-concept demonstration at NASA's Johnson Space Center, the robot was instructed to locate and move a medical bag.

The experiment used the open-source Pi0 vision-language-action model, developed by Physical Intelligence, together with a camera configuration designed to improve spatial awareness. According to Google, the system achieved an initial success rate of 85 percent in these laboratory tests. Although the results demonstrate the feasibility of the approach, they do not establish that the robot is ready for deployment during space missions. Further training, testing and safety validation remain necessary.

Vision-language-action models connect language instructions with physical actions. This could eventually allow astronauts to direct robots using ordinary spoken or written commands rather than specialised programming interfaces.

Computing without relying on Earth

Another challenge involves the computing infrastructure required to operate advanced AI systems. Spacecraft cannot depend on continuous connections to terrestrial data centres, particularly during missions to Mars. AI models must therefore be capable of operating locally with limited energy and computing resources.

Google used its Cloud G4 infrastructure to train and fine-tune the four-billion-parameter Pi0 model. According to the company, this process took approximately 2.5 hours. The researchers subsequently demonstrated that the adapted models could operate on compact, energy-efficient edge computing devices.

Over two years, the team evaluated approximately 15 different devices to identify hardware suitable for environments where power consumption, weight and available space are restricted. The longer-term objective is a robotic assistant that can respond to voice instructions and perform tasks under direct human supervision.

Potential medical applications include preparing diagnostic equipment, assisting with basic physiological measurements and supporting ultrasound examinations. These capabilities remain development goals rather than demonstrated clinical functions.

Space medicine and healthcare innovation

The project also reflects a broader trend in space medicine: developing compact, reliable healthcare technologies that can function where conventional medical infrastructure is unavailable. As we previously reported in an article about medical innovations developed for astronauts, space research has contributed to developments in telemedicine, remote patient monitoring and portable diagnostic equipment.

One example is Tempus Pro, a portable monitoring and telemedicine system originally developed for the European Space Agency. Technology of this kind has subsequently found applications in environments such as aircraft and military operations. Researchers have also investigated robotic medical procedures aboard the International Space Station. In February 2024, the miniature SpaceMIRA surgical robot (in Dutch) underwent an experimental demonstration involving remote manipulation of surgical instruments.

Unlike the proposed NASA-Google assistant, SpaceMIRA was designed to investigate remotely controlled surgical capabilities rather than general logistical and medical support. Both approaches illustrate efforts to provide medical assistance in environments where access to conventional healthcare facilities is severely limited.

Potential applications beyond space

Technologies developed for space missions may eventually offer opportunities for healthcare on Earth, particularly in remote locations, disaster response and other settings with limited medical resources. Compact diagnostic equipment, locally operating AI and robotic assistance could potentially support healthcare professionals where specialist expertise or infrastructure is difficult to access.

However, the NASA-Google system remains an early-stage technology demonstration. Its suitability for actual medical procedures, autonomous operation and deployment in space has yet to be established. The next development stages will focus on improving reliability, refining robotic movements and expanding the range of tasks the system can perform.

For now, the research demonstrates how AI and robotics could help astronauts manage practical and medical challenges during increasingly distant missions, while keeping clinical decisions under human control.