NASA and Google Public Sector are testing an artificial intelligence-powered robotic system designed to help astronauts manage cargo, logistics, and routine medical tasks on future deep-space missions, addressing communication delays of up to 44 minutes during Mars expeditions.
Deep-space exploration demands a level of self-sufficiency. When astronauts travel far enough from Earth that communication signals take nearly three-quarters of an hour to make a round trip, real-time medical guidance from mission control becomes practically impossible, as Jim Kelly, Google’s public sector vice president noted.
To prepare for those isolated conditions under the Artemis program, NASA’s Human Research Program, the Johnson Space Center Robotics Team, and the Jet Propulsion Laboratory teamed up with Google Public Sector to develop a versatile robotic assistant. The project expands on earlier digital tools by giving the system a physical form capable of handling physical payload management and emergency medical support. This current initiative expands upon the Crew Medical Officer Digital Assistant, which was jointly created by NASA and Google to provide astronauts with live medical advice.
Inside the Laboratory Trials
Engineers tested the prototype in early laboratory demonstrations, running controlled “pick-and-place” trials to evaluate how well the machine handles objects. During these initial trials, the robotic system was tasked with locating a medical bag and transporting it to a designated spot.
NASA astronaut, engineer, and scientist @Astro_Christina joined Google’s James Manyika on our latest episode of Dialogues on Technology and Society, to discuss the partnership between astronauts, robotics, and AI, and explore the big question — “Are we alone?”Learn more and…
To navigate confined spaces safely, the Jet Propulsion Laboratory adapted satellite situational-awareness intelligence tools originally designed to track objects in orbit, helping the robot navigate 3D spaces. The development team also relied on Google DeepMind’s MuJoCo physics simulator to assess movement and object handling before attempting more demanding applications.
By pairing an “eye-to-hand” camera configuration with the open-source Pi0 model developed by Physical Intelligence on Google Cloud infrastructure, the robotic system achieved an initial 85% success rate in those early lab environments. Ultimately, the project aims to create a robotic machine capable of performing diverse duties through straightforward verbal commands, operating strictly under the crew’s oversight and direction.
Payload Handling and Medical Support for Lean Crews
Because every minute for a small crew in space is strictly budgeted for complex science and discovery, offloading repetitive labor is a top priority. The machine functions primarily as a payload handler designed to manage cargo, logistics, and equipment.
To maximize its utility for a lean crew, the robotic system is designed as a versatile ‘payload’ handler capable of managing cargo, logistics, and equipment.
Jim Kelly, Google’s public sector vice president
When the robot is not organizing tools, securing compartments, or moving cargo, it can pivot to assist with onboard healthcare needs. For instance, if an astronaut suffers from severe back pain—a potential indicator of kidney stones—the machine could retrieve and position an ultrasound machine.
That physical assistance leaves diagnosis and treatment decisions entirely under human supervision. The human crew medical officer examines the resulting scans and determines the appropriate medical response, ensuring the technology acts as an extra set of hands rather than an autonomous decision-maker.
Edge Computing Challenges and Local Processing Power
Operating outside Earth’s continuous data connection requires running complex models locally on the spacecraft. Using Google Cloud G4 computing instances, the technical staff successfully trained and optimized the 4-billion parameter Pi0 model in a mere 2.5 hours.
Over a two-year period, the team tested approximately 15 different edge-computing devices to identify lightweight, energy-efficient hardware required for future Mars missions. While the current demonstrations prove the concept in a lab, the project remains a proof of concept without a confirmed flight test or space-ready deployment schedule.