MIT engineers have developed a new algorithm that can identify and prevent potential failures in complex autonomous systems. This algorithm is more sensitive and can find hidden correlations, making it more effective than conventional methods. The team demonstrated its success through simulations of power supply grids, aircraft collision avoidance systems, and rescue drones. They also tested it on an actual robotic arm, with results that matched their predictions.