Sophia Elliott 
- NASA’s Human Research Program, Johnson Space Center and JPL are collaborating on the effort
- The system located and moved a medical bag in 85 percent of initial lab trials
- Astronauts would direct the robot through plain-English text prompts
A proof-of-concept robotic system that Google Public Sector is developing with NASA to handle equipment for astronauts located and moved a medical bag in 85 percent of initial lab trials, Jim Kelly, vice president of federal at Google Public Sector, wrote in a blog post Tuesday.
The system is being built under a collaboration between Google Public Sector and NASA’s Human Research Program, Johnson Space Center Robotics Team and Jet Propulsion Laboratory for future deep space missions under the Artemis program and beyond.
What Is the Robotic System Designed to Do?
Kelly, a Wash100 Award winner, said the system is designed as a general payload handler that can manage cargo, logistics and equipment. It can also be used for medical support. In one example Google described, the robot could retrieve and prepare an ultrasound machine so a crew medical officer can perform a scan.
Astronauts would direct the system through text prompts in plain English, using vision-language-action models that link language to physical movement. The long-term goal is a robot that takes voice instructions under astronauts’ direct supervision, potentially helping with routine health monitoring such as capturing basic vitals under a crew medical officer’s guidance.
How Was the System Tested?
To test the system safely at Johnson Space Center’s iMETRO facility, the team used Google DeepMind’s MuJoCo physics simulator, which lets the artificial intelligence practice movements in a virtual environment before working with real hardware. JPL is adapting tools originally built to track objects in orbit to help the robot navigate and understand its 3D surroundings using camera input.
In the pick-and-place demonstration, the open-source Pi0 model, developed by Physical Intelligence on Google Cloud infrastructure, used an “eye-to-hand” camera configuration to find and move the medical bag. Google called the 85 percent result an initial baseline ahead of further training and safety refinement.
Because spacecraft cannot depend on constant links to Earth-based data centers, the team used Google Cloud G4 instances to fine-tune the 4-billion-parameter Pi0 model, which took 2.5 hours, according to Google. The team demonstrated that the fine-tuned models can run locally on compact edge devices. Over the past two years, Google has tested about 15 edge computing devices to identify low-power hardware suited to future Mars missions.
What Other Work Have Google and NASA Done Together?
The robotics effort builds on the Crew Medical Officer Digital Assistant, or CMO-DA, an AI-powered clinical decision support system that Kelly, then vice president of federal sales, detailed in a Google Cloud blog post in August 2025. The tool uses natural language processing, machine learning and a multimodal interface to help crew medical officers or flight surgeons diagnose and treat medical issues when real-time contact with Earth-based experts is limited.
The work supports NASA’s Artemis program, which completed its first crewed flight around the Moon with Artemis II in April. NASA plans an Artemis III demonstration in low Earth orbit in 2027 and its first crewed Artemis Moon landing in 2028, and it is planning a Moon Base near the lunar South Pole.
For more than 50 years, NASA’s Human Research Program has studied how spaceflight affects the human body and developed methods to protect astronaut health and performance.
