Stanford's OWM AI model learns to dock spacecraft with the ISS

Stanford researchers unveiled the Out-of-this-World-Model (OWM), which "dreams" dozens of possible futures to control spacecraft docking with the ISS. It mastered maneuvers in 500,000 iterations versus 25 million for reinforcement learning, though it docked successfully only 53% of the time versus RL's 29%.

• OWM learns physics from experience and predicts outcome probabilities instead of using hard-coded equations
• Training relies on AstroJAX, a library built to run on GPUs
• Docking success across ISS ports reached 53%, compared with 29% for the RL baseline
• The model handled unseen docking ports and unexpected obstacles better than RL