The Mars Polar Lander never had a chance.
At 12:02 p.m. California time on December 3, 1999, after an 11-month journey to Mars, the NASA spacecraft slewed its antenna away from Earth in preparation for entry into the Martian atmosphere. That was the last time mission controllers heard from it. According to the scenario a NASA accident- review board deemed most likely, the Lander dropped out of orbit, deployed its parachute, and began firing its descent engines to slow its fall-just as it was programmed to do. But as the craft’s three landing legs automatically unfolded, sensors in the legs sent false signals to the Lander’s control software, indicating that it had touched down. Not programmed to deal with such a scenario, the software ignored signs that the craft was still aloft and, at an altitude of 40 meters, shut down the descent engines. Gravity took over, and the delicate craft slammed into the rocky Martian surface with the energy of a high-speed car crash.
That same year, but millions of kilometers away, another NASA craft dealt with crisis more adroitly. Deep Space One had just begun a turn to take optical readings that would guide a planned flyby of a nearby asteroid. As it turned off its camera to conserve energy, the power switch stuck open. This redirected power needed by other essential components, interrupting the maneuver and threatening to put the spacecraft into a semicomatose “fail-safe” mode that could have taken ground controllers weeks to undo. By that time, the asteroid would have been left far behind.
But Deep Space One had something Mars Polar Lander lacked: an onboard robot able to think autonomously and handle the unexpected. Using its engineering knowledge, the robot tried to repair the switch by toggling it on and off. When this failed, it devised a successful plan to complete the navigation maneuver, and the craft proceeded unharmed.
The robot that saved Deep Space One was in the vanguard of a new breed of machines poised to have a big impact in space and here on Earth. Quite unlike the metallic contraptions that march stiffly through sci-fi movies or the mindless, stripped-down devices that heft parts on our assembly lines, the new robots have more brain than brawn. Each possesses a detailed picture of its own inner workings-encoded in software-based models-that gives it the ability to respond in novel ways to events its programmers might not have anticipated. Because many of these inward-focused, self-reconfiguring machines don’t move, some computer scientists call them immobile robots, or “immobots.”
Immobots are already beginning to crop up in situations where autonomy is important. They are needed either because direct operator control is impossible (for example, in space probes so distant that radio signals take minutes or hours to reach them) or because humans lack the skill or the desire to oversee all the details (in such down-to-earth systems as office machines, water treatment plants, and internal combustion engines). “There are lots of systems in the world that have sensors and actuators, but don’t look like traditional mobile robots,” says Brian Williams, a former NASA researcher who coinvented Deep Space One’s autonomous software and is now a professor at MIT’s Space Systems and Artificial Intelligence Laboratories.
Once programmed with immobotic software, Williams explains, these systems “have a commonsense model of the physics of their internal components and can reason from that model to determine what is wrong and to know how to act.” Such systems are more self-reliant than typical computers, which are very good at executing mindless, step-by-step instructions laid out for them by software engineers. However, computers are still amateurs when it comes to thinking their way through unforeseen crises such as component failures.
Immobotic reasoning is turning up in everyday systems, including cars, office copiers, and database software. And if advances in the field continue apace, many of the infrastructure technologies we depend on-heating, ventilation, and air conditioning systems; telephone and computer networks; air traffic systems; and electrical grids-might eventually be run by immobile-robot brains. Indeed, entire cities and regions could be transformed into massively distributed immobots. “The problems where you stare at the computer and scratch your head-that’s where sophisticated modeling really has a benefit,” says Sam Lightstone, a senior technical-development manager at IBM’s Toronto Research Lab. “It allows a computer system to make choices and analyses that a human being couldn’t.”