Artificial Intelligence May Aid Space Mission Technical Glitches

SUMMERLAND KEY, Fla. -- NASA officials are reporting encouraging results from an experiment in which artificial intelligence software is being used to determine the root causes of simulated technical glitches aboard the agency's Earth Observing-1 imaging satellite.

The software, called Livingstone, was developed by computer scientists at Ames Research Center in California. The Ames team named the software after the 19th Century explorer and doctor, David Livingstone. The version being tested on Earth Observing-1, or EO-1, is a more powerful version of the Livingstone software that was first tested successfully on NASA's Deep Space 1 spacecraft in 1999.

Livingstone's designers are convinced that artificial intelligence software will be the best way to prevent technical mishaps during future robotic or human missions into deep space. Continuous communications with Earth will be impossible on these missions due to signal-transit delays and planetary obstructions. Spacecraft therefore will have to self-diagnose technical failures, and if possible find a solution, said members of the Livingstone team.

The loss of NASA's Mars Polar Lander spacecraft in 1999 is a good example of the type of failure that Livingstone-type software might someday help prevent, said Sandra Hayden, project manager for the Livingstone experiment at Ames.

The Mars Polar Lander is believed to have crashed when its onboard sensors mistook deployment of its landing legs for actual touchdown on the surface of Mars, she said. The spacecraft turned off its engine prematurely and plunged to the surface.

"The engine was shut off. However, the altimeter would have showed that you were still at some altitude. Something like Livingstone could have diagnosed that conflict," Hayden said.

In a best-case scenario, the craft's computer would have kept the engine running and saved the day, she said.

Generally speaking, NASA officials believe increased spacecraft autonomy will pave the way for new types of missions while saving time and money.

The $500,000 Livingstone EO-1 experiment is the latest in a series of artificial intelligence software demonstrations aboard that spacecraft. In another example, NASA successfully tested software that enabled EO-1 to scan Antarctica's Mount Erebus with a heat-sensitive camera and autonomously focus on the most interesting feature: a lava pool at the summit.

EO-1 was built under NASA's New Millennium technology development and test program and launched in November 2000. In 2002, NASA announced it would extend the mission, and EO-1 has since become "a test bed for evaluating autonomous procedures," said Bryant Cramer, EO-1 program manager at the agency's Goddard Space Flight Center in Greenbelt, Md.

The Livingstone experiment is not as challenging as monitoring all critical systems aboard an operational spacecraft, Hayden acknowledged. For the EO-1 experiment, the Ames team focused exclusively on the satellite's imaging system, Hayden said.

A key feature of the software is its "reasoner" function, which enables it to compare the predicted performance of a spacecraft's systems based on readings from onboard monitors. Contradictions between the predicted and actual performance of the systems are used to identify the root causes of technical failures, Hayden explained.

For the EO-1 experiment, the software was given a detailed computer-code model of the spacecraft's two imaging cameras, Hayden said. Livingstone software can accept models of a variety of spacecraft systems, she noted.

The software was then uploaded to the orbiting EO-1 craft from a control center at Goddard.

Over the last several weeks, Hayden and her colleagues have fed a series of bogus data readings to Livingstone to simulate trouble with the EO-1 cameras. As of mid October, the Ames team had run through 14 of 15 planned failure scenarios, and Livingstone was performing "very well," Hayden said.

In one of the more interesting tests, the Livingstone software correctly diagnosed that a camera cover had failed to open.

Prior to the orbital demonstrations, engineers ran dress rehearsals on the ground using a computerized test bed. "We're happy that we were seeing the same results as in the test bed," Hayden said.

The final failure scenario will be tested in mid-October, but the Ames team will continue operating Livingstone aboard EO-1 through December. "We want to run Livingstone for an extended period of time. That's an important thing to do because it shows it's mature enough to run as flight software. We'll show that we can run quietly in the background without interfering with the normal operations of the spacecraft," Hayden said.

Livingstone is a candidate to serve as the health-monitoring software for the robotic Jupiter Icy Moons Orbiter spacecraft, which will have to operate without contact with Earth for 40 day stretches, said Serdar Ucken, technical leader for discovery systems at Ames.

In the nearer term, the Ames team hopes to convince managers of the international space station to deploy Livingstone as a means of quickly diagnosing equipment problems.

"On the space station today it takes literally hours to figure out root causes of failures," Ucken said.