KEY WEST, Fla -- U.S. satellite software companies are clamoring to fly new artificial intelligence software on a variety of government-owned satellites in the hopes of persuading risk-conscious commercial satellite manufacturers to adopt their products.
Several small U.S. companies have been working on artificial intelligence software since the late 1980s and early 1990s. Their goal is to automate satellite decision-making and orbital planning even more than today. More satellites could be operated by a single control team or staffing could be reduced sharply if these processes were automated with fewer human checks, they said.
Today, when a potential problem arises, the software automatically puts the satellite's electronics in a so-called safe mode so controllers can assess the situation and make adjustments if needed. Ground software automatically plans discreet spacecraft maneuvers, but not without numerous human checkpoints along the way.
The success of NASA's experimental Deep Space 1 science probe earlier this year drew attention to the fledgling artificial intelligence industry, software executives said. During a May experiment, controllers programmed the spacecraft's camera to stay on even after its experimental flight software ordered it to turn off. The software autonomously worked around the mock problem by leaving the camera on and considering the extra power consumption in its new operating plan.
"As you know, the satellite industry is risk averse. [Artificial intelligence] is a major change and it's happening slowly. But it is a growing trend," said Mark Roberts, vice president of business development for AI Solutions, a 35-person company founded in 1992 in Lanham, Md.
AI Solutions, Altair Aerospace Corp. of Bowie, Md., and Interface & Control Systems, Inc. (ICS) of Melbourne Fla., are the leaders among a small group of start-up companies pushing artificial intelligence. Larger, established companies such as Raytheon Co.'s Space Systems business unit in Denver and AlliedSignal Technical Services Corp. in Columbia, Md., also are working on artificial intelligence, sometimes in collaboration with the start-ups.
Each company has its own strategy to prove that its software can save money and improve efficiency without placing expensive spacecraft at risk. The ultimate goal is to break into the commercial satellite industry. So far, however, government-owned research satellites have been the most fertile ground.
For instance, AI Solutions' maneuver planning software will help NASA's Earth Observing-1 satellite fly in tandem with the Landsat 7 satellite so scientists can compare environmental images, Roberts said. Earth Observing-1 is tentatively slated for launch in April.
ICS's control software helps command the telescope in NASA's Far Ultraviolet Spectroscopic Explorer, a company spokesperson said. Altair will soon upload a flight version of its Altairis ground control software for NASA's wounded Wide-Field Infrared Explorer, said Chris Wheal, vice president for marketing at Altair.
Convincing commercial satellite builders to buy into the software has proven harder. Some officials noted that the satellite manufacturers are rightfully cautious at a time when commercial space investments are being closely scrutinized.
"If you're talking to a guy at a small start-up company, it's naturally a culture that's very aggressive. Spacecraft manufacturers are just the opposite. Their lifeblood is reliability," said Mike Ruggles, director of science systems in Raytheon's Denver operations.
Raytheon Eclipse control software is automated but is not artificially intelligent in the strictest sense of the term, Ruggles said. Eclipse is designed to have people in the loop at numerous points. But Ruggles said the company has a research project under way with Altair to explore the possible benefits of greater artificial intelligence. The benefits of reduced personnel need to be weighed against potential risks and costs of additional testing, he said. "That is what we're doing now with Altair. We're trying to answer that question," he said.
To fight the impression that their product is risky and futuristic, software writers avoid using the term artificial intelligence. The words can conjure unwarranted fears of out-of-control satellites ignoring their human builders and acting unpredictably, they said. The software writers prefer to call their products autonomous control systems, expert systems or remote agents, as NASA officials called the experimental software on Deep Space 1.
Beyond semantics, a major debate in the satellite software industry is how much software should be moved to the spacecraft and how much should remain at the fingertips of human controllers at the ground station.
"The typical approach has been to make the satellite very simplistic. More stupid. You leave more to the ground. Our angle is to actually make the satellite smarter," an ICS spokesperson said. Both the ground station and the satellite operate with the same software. "That's a tremendous advantage because it's standardized."
Some controllers want the autonomous software on the ground. "You're not up there with the bird. There a perceived risk in shifting that on board," said Mike Fatig, director of space services for AlliedSignal Technical Services.
Nevertheless, Fatig said coupling ground software with space software is a growing trend that his company is highly involved in. "We're a ground system provider but we are getting involved more and more in flight software. With the addition of flight software, you can actually change a mission in orbit," Fatig said.
Wheal predicted that autonomous software tools will prove to be more reliable in the long run. "They don't send the wrong commands," he said.
