Researchers are combining neural networks and robotics for a new breed of small satellites.
NASA technicians at Johnson Space Center’s didn’t believe Mark Tilden was serious. The prototype satellite he wanted them to test could almost fit in Tilden’s pocket. The device weighed only a few ounces and was far and away the smallest spacecraft ever analyzed in NASA’s five-story vacuum chamber. But it passed its tests with flying colors and is now slowly making its way through additional testing and development at Los Alamos National Laboratory, where it was originally conceived and designed. Its inventors have a one-way trip to outer space planned for it eventually.
The tiny satellite, along with the other spacecraft being developed at Los Alamos, differ from traditional satellites in more ways than just size. They’re also robots, an outgrowth of work by Tilden, a roboticist; Jan Frigo, a senior research engineer with expertise in adaptive control systems; and Kurt Moore, an aerospace scientist.
STARTING WITH ROBOTS
Much of the Los Alamos research centers on the design of autonomous robots controlled by artificial nervous systems (a nervous net, Nv, in Los Alamos parlance), rather than by microprocessors, memories, and software. The basis of these Nv control systems is a two-transistor “motor neuron” that produces control pulses proportional to the perceived load on an inductive actuator or sensor. As in biological nervous systems, neurons can be linked together, with the output of one serving as the input for the next, and so on. When a series of these nonlinear neurons are configured in a bounded, phase-locked ring, they can mimic biological pattern generators and serve as controllers for multilegged, walking robots.
To make the robots as autonomous as possible, the design team uses solar cells for power. But current solar cells are inefficient and can’t provide enough electricity to continuously power a motor system. So the robots store energy in capacitors until there’s enough to do some useful work.
The team’s experience with over 200 Nv robots has convinced them that such simple systems are extremely reliable and can often show surprising emergent behavior, behaving in ways that were neither planned for nor predicted. Nv robots can also handle electrical and mechanical failures that would paralyze robots controlled by traditional processors.
“You can pull all but one leg off some of our walking robots and they will learn to crawl along on that last remaining leg like pathetic robot roadkill,” notes Tilden. “We also have robots that will fight each other and then suddenly learn that cooperation and coordination are better if any of them are going to move around at all.”
Tilden recently built a series of structurally identical BEAM ants. (BEAM stands for biology, electronics, aesthetics and mechanics, the principles behind Nv robots.) The ants differed only in the placement and length of their tactile sensors, which affects the controller’s oscillator. “We put them in the sun and used time-lapse photography to observe them,” says Frigo. “Over time, they exhibited a kind of blocking behavior, forming a V. Those having the longest sensors wound up out front.”
But though the robots can display positive emergent behavior, such as self assembly and teamwork, they can also show pathological traits as well. “We have a lot of first-order machines that we call very good bad examples,” says Tilden. “They fall into pathetic forms of epilepsy or when they get confused, they roll on their backs and kick their legs in the air. We try to improve robot survival instincts by minimizing deficit behaviors while increasing positive behaviors.”
Over the years, Tilden and his team have built increasingly sophisticated robots. “At one end, we have machines that can barely pull themselves toward the light,” explains Tilden. “And on the high end, we have devices that can solve the Indian Monkey Trap conundrum. That’s where fruit is in a glass bottle having an opening large enough to let a monkey get his hand in, but too small to let him get his hand out without letting go of the fruit. The monkey sees he has the fruit in his hand and refuses to let it go, so hunters can walk up and kill him. Our robots, after trying different methods to get at the target, will give up on that one and go for another.”
The Los Alamos team has added secondary back-up controllers that detect frustration characteristics and change the robot’s behavior. A robot designed to seek light, dubbed a photovore, might try to get through a slit too small for its body, for example. A first-order robot might just get his head stuck in the slit. Those with back-up controllers, however, might back up, try a second route to the left or right, do stochastic wall following, or try a variety of other behaviors. “Our most recent creatures do all this, and most interesting, some try to dig a tunnel to the light.”
In general, the team has abandoned the traditional idea of designing robots or systems to perform a specific function then trusting them to survive all anticipated and unanticipated circumstances. Instead, the Los Alamos paradigm is to first design systems that inherently try to survive all circumstances, then try to get them to do useful work. Outer space is one place the Los Alamos team sees applications for such robots.
TRAINING FOR SPACE
The Los Alamos research dovetails nicely with NASA’s move in recent years to “faster, cheaper, smaller” space programs. The Los Alamos robots are so small, however, they needed a new term to distinguish them from NASA’s microsatellite concept. Microsatellites, says NASA, weigh less than 660 lb. Nanosatellites, such as those being developed at Los Alamos, tip the scales at 22 lb or less.
The first nanosat, a prototype, will use three torque coils, one per axis, to orient itself in the Earth’s magnetic field and a photosensor pair to point itself at the brightest available light source. The three coils or magnetic torque stabilization vanes, are not reliable for large masses, so the nanosat will weigh in at 50 gm or less.
The low weight also contributes to power efficiency. The plan is for small solar-cell arrays to provide power. The nanosat would rest while storing energy, then gather or transmit data, or reorient itself. “Current prototypes use a half watt for initial orientation, then operate with only 0.01 W of standby power after the target is acquired,” says Tilden.
The team envisions several different methods of getting nanosats into orbit. The spacecraft could be piggybacked on the launch of a larger, more conventional satellite or taken up on a Shuttle mission. They could also be put in orbit on a Pegasus-like missile fired from a high-altitude aircraft. Ideally, though, hundreds of nanosats would be launched simultaneously on a dedicated rocket.
Although the first prototype will lack self-contained propulsion, future versions probably will not. “Right now we’re assuming simple polar-synchronous orbits where the devices will get enough power to turn and communicate, but won’t be able to translationally move themselves.”
“In the future, we would want to be able to take a nanosat out of orbit once its missions was over or if it failed catastrophically,” says Frigo. “We wouldn’t want to clutter up space. But for now, propulsion isn’t part of our mission or goals.”
The team is now focusing on improved power efficiency, noise rejection, and pointing accuracy of nanosats. “Currently, we can only get accuracies down to 2°, but we’re hoping to improve that with better sensors and improved machining,” notes Tilden.
The obvious advantage nanosats have over traditional satellites is their small size and weight. With launch-to-orbit costs still in the $3,000/lb range, smaller spacecraft cost less to get into orbit. Relying on numerous redundant nanosats instead of a single large satellite gives a space mission built-in fail-safes so it can continue even if half the nanosats are destroyed. Smaller satellites are also better suited to certain missions. Gathering data on magnetic fields, for example, is difficult for satellites so large they create their own magnetic fields.
Nanosats are also particularly well-suited to exploring the Van Allen radiation belt and other high (greater than 1,000- km) orbits. Because they carry no microprocessors or delicate electronics, they are relatively immune to radiation. Traditional satellites orbiting far from Earth, such as GPS spacecraft, must be radiation hardened, an expensive proposition.
But small satellites have a downside: they’re are susceptible to damage from collisions with space junk. “There’s already quite a lot of debris in orbit from past space missions,” says Frigo. “A nanosat could be destroyed if it were to get hit by any of it, so collision avoidance would be a priority.”
There are several engineering problems that must be solved before nanosats become viable information gatherers. One of the most important is figuring a way to get the information from the nanosats to Earth without burdening the small spacecraft with the weight of a large radio transmitter or a power source to run it.
The Los Alamos team has several possible solutions. Nanosats designed to seek out magnetic minimums or maximums, or some other celestial parameter, might not need transmitters. They could be tracked from Earth using radar or lasers. “It would be as if you were tracking Ping-Pong balls on the surface of the ocean,” explains Tilden. “It would give a good picture of the sea state in the particular region.” He points out that a three-torque-ring nanosat closely resembles a cornercube, an object that shows up brightly on radar. And nanosats could modulate information onto laser beams that bounce off of them. “NASA’s Jet Propulsion Lab recently managed to bounce a laser off a low-Earth-orbit satellite and get active reflective characteristics back,” notes Tilden. “But using lasers would be difficult,” he admits.
Another method is to equip a flock of nanosats with small transmitter/ receivers and have them talk to each other while gathering data. That data would be transmitted to a larger mothership stationed near some point in the flock’s orbit. The mothership would then download the data to Earth. The final possibility is to have each nanosat in the flock send a string of pulses at the same time. The combined signal would be strong enough to be picked up by phase-array antenna on Earth. “These are all techniques that will be addressed much later in the research,” says Tilden.
Another aspect that will have to wait is the design of mission payloads small enough to fit aboard a nanosat. A CCD camera, gamma-ray detectors, and microradio antennae are all on the planning boards.
One of the first missions researchers have in mind is exploring the Earth’s magnetopause, the outer reaches of the magnetic field surrounding the planet, and how the Sun affects it. “We’ve never been able to do real-time monitoring of the magnetopause position as the solar wind pushes it around,” says Moore. “With hundreds of these microsatellites, we should be able to that. And they could use torque coils to both orient themselves and detect magnetic fields, while information could be stored and transmitted to Earth.”
Nanosats might also map variations in the Earth’s magnetic field. “This type of mission would be worthwhile because the magnetic field affects many expensive, high-quality satellites,” notes Tilden. “And scientists need better predictions of magnetic turbulence and the extent of the magnetopause. It’s a relatively simple project and a good first task to prove nanosat technology.”
Another novel space mission Tilden and his team foresee is the exploration of passing meteors or asteroids. They would equip simple nanosats with pyrofors, devices developed for NASA that generate sparks when they hit something hard. A shotgun-blast of several hundred nanosats would be launched at the passing object, with the shape of the dispersion known. As the kamikaze nanosats hit the surface and destroy themselves in a bright flash, scientist could record the flashes, using time-based recordings to derive topological data and spectral analysis to determine the object’s elemental make up.
Testing and development for space is expensive and time consuming, and the Los Alamos team doesn’t expect to fly its first nanosat for three to five years. “We’re trying to get time on NASA’s KC-135 vomit comet, the aircraft used to create a few seconds of zero gravity while in flight, to get some microgravity tests,” says Tilden. “Then, after some more testing, we’ll be able to put down some hard specs for the first nanosats.”
“These microsatellites can go where expensive, big satellites can’t, and they can perform a class of business and science missions no other platform can,” adds Moore. “Nobody knows how little you can make them, but we aim to find out.”
A ROBOT FOR EVERY NICHE
“There are 110 million mines out there killing one child every 14 minutes in 63 different countries,” says Tilden. “It’s the worst pollution problem in terms of human suffering bar none.” But in a case of unanticipated results, the team of people hired to clear the minefield didn’t feel right about sending a robot on a oneway mission. “It was 60 transistors and $215 worth of hardware designed to save their lives, but they didn’t want to use it because they felt sorry for the robot.”
In collaborating with industry, he has worked on oceangoing aquabots built to seek out geologic deposits and dreams of “evolving” them into “shepherd dogs” that would herd fish for harvest. Another aquabot variation would listen for the sounds of fishing boats in areas off-limits to fishing and then broadcast a signal that scares away fish, thus making it uneconomical to fish illegal waters.
Another task he would like to see his creations performing is cleaning up after nuclear accidents. “There were 28 robots from different countries sent to Chernobyl to help the cleanup,” notes Tilden. “They all died. Radiation fried their processors regardless of how much rad hardening they had. Ours are inherently rad hard since they’re analog.” He envisions a herd of small robots designed to scurry around collecting radioactive debris and depositing it in a central waste dump. When there’s no more to find, they’d hurl themselves on top of the pile.
For more information on Tilden and his robots, including his annual robot competition, check out his Web page at ssl.lanl.gov/robot