Spacecraft docking model beats comparison system in simulation

Stanford researchers report that an AI model docked successfully at about 53% of space station docking ports in simulations, compared with 29% for a reinforcement learning system. Their preprint says the model also handled unseen docking ports and an obstructed target more effectively. The system learns from simulated flights and tests possible approaches before choosing its next control action. Researchers trained it using graphics processors; they report 500,000 iterations for the model versus 25,000,000 for the comparison system. Its 53% success rate leaves it far from reliable spacecraft deployment.