THE ESSENTIALS IN 30 SECONDS

  • NASA's Human Research Program, Johnson Space Center robotics team and JPL worked with Google Public Sector on the proof-of-concept system.
  • The robot uses Physical Intelligence's open-source Pi0 model, with 4 billion parameters, fine-tuned on Google Cloud G4 instances in 2.5 hours.
  • In early lab demonstrations it reached an 85% baseline success rate at locating and moving a medical bag; no flight tests have been announced.

Google Cloud and NASA are testing a robotic system designed to help astronauts with routine chores and medical situations on long missions, when help from Earth may be too slow to arrive. Google described the work in a blog post on October 27, and NASA confirmed to Nextgov/FCW that it is assessing the tools.

A robot for cargo and first aid

The system was developed by NASA's Human Research Program, the Johnson Space Center robotics team and the Jet Propulsion Laboratory, together with Google Public Sector. It is presented as a proof of concept connected to the Artemis program, which aims to return astronauts to the Moon and later send crews to Mars.

The robot is primarily a payload handler for cargo, logistics and equipment, with tasks such as organising tools or securing compartments. It can also step in for medical support. Google's example is fetching and preparing an ultrasound machine for an astronaut with severe back pain, a possible sign of kidney stones. The project extends the Crew Medical Officer Digital Assistant, a clinical decision support tool NASA and Google introduced earlier. Google cites round-trip communication delays of up to 44 minutes on a Mars mission as the main reason to give crews more autonomy.

How it was built and tested

The robot runs Pi0, an open-source vision-language-action model from Physical Intelligence with 4 billion parameters, which lets astronauts give instructions as plain English text. According to the reports, Google fine-tuned the model on Google Cloud G4 instances in 2.5 hours and used Google DeepMind's MuJoCo simulator for virtual testing. The fine-tuned models were shown running locally on small, low-power edge devices; over two years the team tried around 15 such devices.

In a demonstration at Johnson Space Center's iMETRO facility, the robot performed pick-and-place tasks to find and move a medical bag, reaching an 85% baseline success rate. The number of trials was not given.

Limits

These were early laboratory demonstrations, not flight tests, and neither company has given a timeline for use in space. Google says the system is meant to act under astronaut supervision rather than replace human decisions, with the longer-term goal of handling a range of tasks from simple voice commands. A NASA spokesperson said work with several partners to evaluate such tools is ongoing.