A toaster-sized robot has taken a major step toward autonomous space navigation.
Researchers at Stanford University have successfully demonstrated a machine learning-based control system for the first time on the International Space Station (ISS).
This achievement marks a turning point in on-orbit robotics, opening the door to missions in which robots operate with minimal human oversight.
Astrobee, a cubic free-flying robot already on board the ISS, served as a test platform.
The new system will allow astronauts to safely navigate the station's narrow corridors and cluttered interior without direct control.
This research was presented and presented at the 2025 International Space and Robotics Conference (iSpaRo).
Smarter movement planning
The ISS environment is dense and interconnected, filled with storage racks, computers, wires, and experimental hardware.
That complexity makes motion planning difficult, said lead researcher Somrita Banerjee, who completed the study as part of her doctoral program at Stanford University.
Traditional planning approaches used on Earth do not translate well to space hardware.
“Flight computers running these algorithms are often more resource-constrained than ground robot computers,” said lead author Marco Pavone.
He added that space has more uncertainties and stricter safety requirements than ground-based robots.
To address this challenge, the team built an optimization system that uses sequential convex programming to plan safe and feasible routes. However, solving each step from scratch requires significant computational time, slowing down the process.
So the team trained a machine learning model based on thousands of previous solutions.
This model acts as a “warm start,” providing an informed first guess before optimization refines the path.
Safety constraints remain, but AI dramatically accelerates the process.
Banerjee likened this to choosing a route based on common travel paths, rather than drawing a theoretical straight line between two cities.
“We start with what we experience and optimize from there,” she said.
Before arriving at the ISS, the system was tested on a floating robotic platform that mimics microgravity at NASA Ames Research Center.
When testing began on the ISS, the astronauts only did some preparation and cleanup before exiting.
The ground team then issued a command through NASA's Johnson Space Center.
The team tested 18 trajectories, running each twice. The first was a standard cold start, and the second was an AI-powered warm start. The results were conclusive.
“We found it to be 50 to 60 percent faster, especially in more difficult situations,” Banerjee says.
These included complex maneuvers that required tight spaces and rotations.
Preparing for future missions
NASA currently designates the system at Technology Readiness Level 5, indicating that it will work in a real-world operational environment. This reduces the risk of future proposals and experiments.
Looking ahead, Banerjee said autonomy will become essential as space missions expand.
“As robots move farther from Earth and missions become more frequent and less costly, we can't always control them remotely from the ground,” she said.
Pavone's team plans to push this technology even further using modern language tools and powerful AI models similar to those behind self-driving systems.
