Remote Agent does the Job for NASA

It was one small step in the history of spaceflight and one giant leap for computer-kind. Remote Agent became the first artificial intelligence to be given primary command over a spacecraft. For two days, starting on May 17, Remote Agent was given primary control of NASA's Deep Space One mission with its futuristic ion propulsion engine. Engineers even simulated several problems which Remote Agent had to decide the best way to deal with. Everyone involved was impressed with the results.

The concept for Remote Agent is yet another step in NASA's "faster, better cheaper" motto. The artificial intelligence will pave the way for future robotics explorers to foam the solar system with minimal ground control. Remote Agent was a high-priority request for technologies that would meet the goals of the New Millennium Program. The New Millennium program calls for NASA to have a "virtual presence" in space. This would involve many small, robotic, inexpensive spacecraft and rovers exploring all parts of our solar system to gather information. To achieve this, NASA requested technologies that would enable spacecraft to control themselves automatically, yet were relatively easy and inexpensive to build. Researchers at NASA's Ames Research Center and Cal Tech's Jet Propulsion Laboratory (JPL) combined forces and rose to this challenge with the development of Remote Agent.

Remote Agent is made up of three components which each play significant, integral role in controlling the spacecraft: The Planner and Scheduler (PS) produces flexible plans, specifying the basic activities that must take place in order to accomplish the mission goals. The Smart
Executive (EXEC) carries out the planned activities. And the Mode Identification and Recovery (Livingstone) monitors the health of the spacecraft and attempts to correct any problems that occur.

The three parts of Remote Agent are constantly in contact with each other as well as the remote components on the spacecraft (cameras, thrusters, instruments etc.) Livingstone receives information about various components onward the spacecraft by way of monitors located at various places. Livingstone knows what a "healthy" spacecraft is supposed to look like. If it detects a problem, it must then diagnose what the most likely cause of the problem is, and then suggest a "treatment" to the Smart Executive. If the problems one which cannot be fixed, Livingstone must report this to the Smart Exec which will then work with the Planner Scheduler to devise an alternative plan to reach the mission's goals.

To test how well Remote Agent could handle problems, engineers programmed in several simulated failures aboard the spacecraft. The failures were correctly handled each time. The simulations were: 1) a failed electronics unit, which Remote Agent fixed by reactivating the unit; 2) a failed sensor providing false information, which Remote Agent recognized as unreliable and therefore correctly ignored; and 3) an attitude control thruster (a small engine for controlling the spacecraft's orientation as needed) stuck in the "off" position, which Remote Agent detected and compensated for by switching to a mode that did not depend on that thruster.

Once the test objectives were successfully achieved, the mission team resumed control of the spacecraft by sending a prepared command to stop Remote Agent and return to ground based control. Remote Agent will be used more frequently on future missions with the ultimate goal that
it completely handle an entire mission unless it runs into a problem for which it must contact ground control.