During the week of May 17, 1999, an AI system called Remote Agent took control of NASA’s Deep Space 1 probe and ran it without human commands from the ground. Built by engineers at NASA’s Ames Research Center and the Jet Propulsion Laboratory, it is widely cited as the first artificial intelligence software to command a spacecraft. The experiment ran for about 29 hours starting Monday May 17, and again for six hours the following Friday, with the team reporting that they met 100 percent of their objectives.
Remote Agent was not a single neural network but a package of classic AI techniques: an onboard planner and scheduler that worked out what the spacecraft should do, an executive that carried the plan out, and a model-based diagnosis system that watched for faults. Rather than following a fixed, pre-uplinked command sequence, the software was given high-level goals and figured out the detailed steps itself, then monitored hardware and replanned when things went wrong.
The experiment deliberately tested that fault handling. At one point the system was told to fire the ion engine and then turn it off, but a timing error meant the engine did not shut down on schedule. Remote Agent detected the discrepancy and formulated a new plan that worked around it. Earlier, when an electronics unit appeared to fail, the software recovered it by reactivating the unit, much like rebooting a frozen computer. Remote Agent went on to share NASA’s 1999 Software of the Year award, and its planning-and-diagnosis approach influenced later autonomous systems on Mars rovers.