Researchers at Stanford University are bringing science fiction visions to life on the International Space Station (ISS).
They demonstrated for the first time that AI-based controls can autonomously guide a robot in space. The research focuses on Astrobee, a cubic-shaped, fan-powered robot designed to float through the narrow, instrument-filled corridors of the ISS.
The system will allow Astrobee to navigate complex modules, avoid obstacles, and perform tasks such as leak detection and supply delivery, potentially freeing up astronauts' time and opening new avenues for robot-powered space exploration.
“We show that robots can move faster and more efficiently without sacrificing safety, which is essential for future missions where humans won't always be able to guide robots,” lead researcher Somrita Banerjee, who conducted the study as part of her doctoral program at Stanford University, said in a statement.
AI guides Astrobee
This result is an important step in space robotics. Traditional autonomous planning methods used on Earth are often impractical due to limited onboard computing resources and stringent safety requirements.
“Flight computers in space are far more resource-constrained and have far greater uncertainties, disturbances, and safety demands than ground-based robots,” Marco Pavone, an associate professor of aerospace and director of the Autonomous Systems Institute at Stanford University, said in a statement.
The research team has developed a route planning system for Astrobee, the ISS's robotic assistant, that utilizes sequential convex programming, a method of breaking down complex trajectory plans into smaller, more manageable steps while ensuring safety and feasibility. However, solving each step from scratch can be computationally intensive and time-consuming.
To accelerate this process, the team integrated machine learning-based models trained on thousands of previous path solutions. The model identifies repeating patterns in hallway layout and obstacle locations, providing a “warm start” for the optimizer. This approach allows Astrobee to generate safe and efficient trajectories faster without compromising constraints. “Using a warm start is like planning a road trip on a route that others will drive before you develop speed and efficiency,” Banerjee said in a statement.
The team says the integration of optimization and machine learning will be the first example of AI-assisted robot control on the ISS, enabling faster, safer, and more autonomous operations essential for future deep space missions.
autonomous space robotics
Before deploying the AI system into space, the team trialled the AI system on a dedicated NASA testbed designed to simulate partial microgravity. The Astrobee-like robot floated above a granite surface, supported by compressed air to mimic the frictionless conditions of orbit. This setup allowed the AI to navigate while encountering virtual obstacles, allowing it to refine its movement plan without risking a collision.
During the ISS experiment, astronauts performed only minimal preparation and cleanup, and the AI system operated autonomously for four hours under remote supervision. Ground operators relayed commands to the robot, specifying start and end points, obstacle configurations, and tested both traditional “cold start” plans and AI-assisted “warm start” plans. Each trajectory lasting more than 1 minute was run twice to allow a direct comparison of the efficiency of the plans.
The results demonstrate that warm start planning can significantly speed up trajectory calculations, reducing planning time by 50-60%, especially in complex environments such as narrow corridors, cluttered spaces, and rotating maneuvers. Multiple safety measures such as virtual obstacles, backup hardware, and abort functionality ensured risk-free operations. After the experiment, the AI system achieved NASA Technology Readiness Level 5, indicating successful testing in a real operational environment.
The researchers say advances in safety-focused, mathematically-based autonomy could pave the way for robots to operate more independently, allowing autonomous systems to handle mundane or dangerous tasks while freeing astronauts to focus on higher-priority tasks and supporting manned missions to places like the Moon and Mars.
Details of the Staford team's research are available on arXiv.
