The first stage of the work involved robust electronic "creatures" that could crawl, slither and exist in such a lifelike way that visitors to the lab averse to real bugs and spiders were loath to handle them. In Hugin, the system will allow the Swedish satellite to track the position of the sun — important because it will supply the energy the craft will rely on to recharge its batteries.

This field of analog robotics was pioneered by Mark Tilden, Brosl Hasslacher and their Los Alamos colleagues, who began by copying simple neural networks from animals such as lobsters, fish and insects. Built from simple analog electronics, sensors and actuators, the artificial creatures were not equipped with what would normally be called "intelligence." They contained no specific knowledge of the world, nor were they explicitly equipped to learn about it. Instead, the networks provided a way for the machines to adaptively interact with the world, and the results have been strikingly lifelike.

Prototype of a three-axis biologically inspired control system that orients satellites toward the sun.

According to Kurt Moore, a Los Alamos staff scientist, the work holds important lessons for engineers. The main advances, he said, "concern minimalist devices that are remarkably capable and that can demonstrate emergent behavior. Although attention has been focused on the controllers, I think [another] lesson is that a better way to design systems is to abandon the 'integration of subsystems' approach and instead design synergistic systems from the start."

Rather than a centralized intelligent control system, the design philosophy is to have an array of controllers accept and process sensor information and produce control outputs for actuators — primarily motors — in an efficient and reliable way.

Reliability comes from the simple but powerful way in which the systems are wired. In the neural architecture, the whole system, including actuators, sensors and controllers, is laid out in a ring structure. As events occur in one part of the circuit, the effects are felt a short time later in all the others. Thus, the input and output of each actuator — which could be an arm, leg or other moving part — are nonlinearly coupled together: The more force a limb experiences, the more it pushes back. The controller neurons used are analog units that "fire" a signal in proportion to some nonlinear function.

The hardware involves only a few basic components — transistors, capacitors, resistors — but the behaviors they create are complex. Partly, the simplicity results from the generation of an internal pattern or rhythm that drives the creature's steps. Oscillations are what give any creature, biological or artificial, a time signature that drives its movement patterns. In some animals, these control impulses are produced by "central pattern generators" located in the spinal cord, and the rhythm allows animals to keep on walking, flapping or hopping once they have made the decision to start.

In the artificial network, the central oscillation both drives, and is driven by, the interactions between the sensors and actuators and the world. The resulting dynamic system has more in common with a fluidic system than conventional digital-processor-based control systems. In a conventional computer, information is sent around in bits or bytes that are independent of their neighbors. But with analog networks, there is no such thing as an independent piece of information, since everything is coupled together.

Because of that interdependence, there is no reliance on "communication" in the conventional sense. Likewise, complicated "sensor-fusion" algorithms to combine incoming data from different sources are unnecessary.

Creatures come cheap

"Like living things," said Moore, "[Tilden's] devices twitch rather than move smoothly. They degrade gracefully in the presence of failures." While the concept of a central, random-movement generator could be patented, in Moore's view, the overall system strategy is the strength of Tilden's creations.

Initially, Hasslacher, Tilden and their colleagues saw their electronic bodies as platforms that could carry other types of artificial intelligence. Even if the higher-level competence was missing, disabled or dead, the machine itself could still survive as long as it was it was not too badly damaged.

However, they also found that the long-term survival of the machine need not be as important as its simple ability to get the job done, particularly because the creatures were cheap enough to be considered disposable for some applications.

An example of this is a project the team carried out at the government's Yuma proving grounds in Arizona, an area that has been littered with mines, missiles, shells and bombs since the 1940s. To clear that section of desert of unexploded ordnance (known as UXO), the scientists designed many different types of "biomechs" to detect and gather the dangerous refuse.

The job was particularly challenging because of the extreme conditions. The temperature could reach 165F during the day and dip to freezing at night, and the nominal humidity was 80 percent with the air full of dust. In addition, there were also real creatures — such as snakes and scorpions — that the machines would have to deal with.

The experiment was not entirely successful, in that the machines tended not to "live" very long. But the fact that they could survive at all, and detect and gather some of the UXO, was encouraging.

Though the robotics work continues, it is a second line of research that has led to the simple sensor-actuator coupling in the new satellite controller. That work has been pursued by Tilden with Moore and another colleague, Janette Frigo. In one experiment, for instance, the network idea was adapted to allow infrared tracking using two sensors coupled to a motor. Each sensor was linked to a neuron that has natural oscillation, the time constant of which was modulated by the photocurrent. As the latter increases, the on-time duty cycle of the neuron decreases. This means that more power is diverted to moving the actuator in the direction of that sensor. This continues until both sensors see the same brightness of target and the system stops moving.

The latest incarnation of that kind of technology is to be incorporated into the Hugin satellite, due for launch on Aug. 15. The main body of the craft was designed by the Swedish Institute of Space Physics and is identical to an earlier craft. However, the control systems for Hugin were designed at the Royal Institute of Technology in Stockholm, with a heavy emphasis on non-standard technologies like neural networks.

In Hugin, 12 photodiodes are initially to be used to detect UV radiation from the sun in order to monitor the orientation of the spacecraft — they will simply represent a source of information for the main controller via an analog-to-digital converter. However, the sensor circuit is also equipped with torque coil actuators and controllers, so that it can be used as a backup attitude-control system, maintaining power through solar recharging despite any other system problems.

If the experiment is successful, then engineers may more often consider the possibility of using cheap, robust but specialized analog systems that can take over if their all-purpose computer controllers fail.

On the implementation issue, Moore agreed that the controller could be done as a microprocessor. However, he said, to do so would be inefficient from a system perspective , increasing the weight of mobile systems. The group had so little funding that it was much easier to build analog controllers from salvage bin parts than microprocessor-based systems.