When NASA launches Deep Space 1 a year from now, scientists on the ground will be able to spend a week with their feet on their desks spectating while the craft flies solo, navigating its way around a part of the solar system and fixing a pre-programmed problem. This is when a freshly minted batch of code will get a chance to earn its wings.
The code, a project called Remote Agent, is an artificially intelligent software system designed to help the craft autonomously manage flight plans, execute them, and correct the flight system when a breakdown occurs – all with little or no interruption to the mission. The mission, DS1, will give scientists including Pandurang Nayak the first opportunity to see how their technology will manage under the rigors of a real space mission – a tough feat during austere times for the space agency.
"It's a Catch-22. Project engineers are very conservative and want only the technologies that have been tested in space," explained Nayak, deputy lead for the Remote Agent experiment at NASA’s Ames Research Center. "But to get that experience, you have to get your technology on a mission."
To help projects like Remote Agent cut their teeth, NASA devised its New Millennium Program, a series of flights, each with different technological goals, that will function as experiment labs for new technologies. DS1, the first of these missions, will venture out to the asteroid McAuliffe, on to a comet and, later, Mars. Remote Agent, the work of five scientists from NASA's Jet Propulsion laboratory and 13 from the agency's Ames Research Center, is one of 12 new technologies that will be tested during the flight.
Remote Agent is a necessary step for NASA, particularly in light of what would seem to be opposing goals - establishing a virtual presence in space while working with an increasingly smaller budget. To achieve this virtual presence, NASA will eventually be launching one craft every month, said Nayak. "The only way to do this is to use technology to decrease the cost and enable more capable missions," he noted.
This capability is achieved through the automation of missions - the forte of a piece of software like Remote Agent.
Building upon lessons culled from missions as recent at the Mars Pathfinder and dating back as far as Apollo 13, Remote Agent is intended to cut down on the large time lapses during the design and transmission of instructions from the ground crew to a spacecraft – delays that can jeopardize parts of missions – while taking over some of the more grueling aspects of unmanned missions, particularly the painstaking monitoring, analysis, and planning often performed by the ground crews.
Remote Agent is made up of three components - a planning and maneuvering module, an execution component, and a fault protection/operation system. At the heart of this system is a model-based artificially intelligent engine that gives Remote Agent a set of possible actions for all aspects of space flight and scientific operations. All three modules essentially allow a spacecraft to think on its feet and potentially eke more scientific research out of a mission.
For example, on Mars Pathfinder, the Sojourner rover carries out science only as it is directed by the ground crew. To come up with a plan for action, the crew has had to look at terrain photographs, decide on an area for investigation, and then determine a path for Sojourner to take and what experiments it should perform. Finally, the crew has had to transmit these instructions within a specific window of time – missed deadlines have cost them an entire day of scientific observations.
But if Pathfinder had Remote Agent's planning and maneuvering module, both the lander and Sojourner would not have to depend up on the ground crews for their daily instructions. The planner operates under the same innate logic that humans do when performing tasks that they may take for granted, said Kana Rajan, senior research scientist at NASA's Ames Research Center. "There's a set of logic that governs how we walk – we don't think much about it," he explained. "But you don't lift both legs at the same time. Something in the brain controls this, and this is similar to how the planner works."
The second module, the executive, sets the activities in motion and tracks whether they've finished, completing the loop of communications within a spacecraft.
Another crowning achievement for Remote Agent could be what it saves in terms of time and opportunity when mission operations don't go smoothly. Through what Nayak calls a "fault operational" module, Remote Agent will detect, diagnose, and fix a craft's problems – without intervention from ground control crews. Normally, when systems fail on a craft, much of its operations shut down, save for communications with ground control, while awaiting instructions.
By contrast, Remote Agent's fault operational mode does a brute-force search through a number of logical statements to come up with the best possible fix, in much the same way that IBM's Deep Blue routed though an index of possible chess moves to respond to Kasparov's strategy. In the process of doing this, a craft doesn't have to shut down, and the solution is grafted into the already-running schedule. The technical difficulty amounts to nothing more than a hiccup.
And it keeps the human role for such missions down to merely looking over the shoulder of a craft. "You don't have to have a human being joysticking or baby-sitting the spacecraft, and this will bring a big savings to missions," said Rajan, who noted that DS1 is capped at US$138.5 million.
