Creating a robot colony
INEEL scientists mimic biological societies with swarms of mini-robots
The mini-robots weave their way through cluttered and dark rooms to find a contaminated spill.
Robotics and Human Systems Research—Scientists at the Idaho National Engineering and Environmental Laboratory are creating an army of small robots--a fleet of inexpensive mini-robots designed to work harmoniously to perform tasks too hazardous or just downright boring for humans. But getting one hundred robots to work together isn't exactly straightforward and the INEEL scientists are taking some inspiration from nature. Simple biological societies, such as ant colonies and beehives, serve as handy models for creating large groups of small, disposable robots.
How nature does it
For years now, robotics engineers and computer scientists have worked toward creating robots that work in concert. If one robot can be sent to search for landmines, think how much faster a group of ten could sweep the area--or one hundred. Plus, searching for landmines is risky. It's good to be armed with more than one robot for the task.
But engineers struggle with the logistics of managing a large group of moving machines. It is not at all economical to assign one operator to each robot. One hour of robot-work would equate to one hundred--or more--man-hours. On the other hand, one human operator managing 30 remote-controlled robots at once doesn't work either.
Communicating with each robot just isn't very practical. "For real robots, there are some serious communication problems," said Donald Dudenhoeffer, one of the lead scientists on the project. "First, getting enough bandwidth; second, communications always go down—always; and third, the line of vision has to be clear. They have to see each other in order to communicate."
So the scientists at INEEL have turned to biology for inspiration. Look at a flock of birds, they say. The birds stay together in a type of fluid order, as if there were a larger force directing the group. But there is no master control; flocking works because each bird responds to its neighbors.
In nature, such large global behaviors emerge out of small individual behaviors all the time. Fish school, birds flock, butterflies migrate. None of these actions are globally coordinated. There is no lead anchovy with a megaphone.
Instead, each fish cues off a few of its neighbors—and the whole group works in harmony. It's the same for ants. Each ant follows the pheromone trail of another one to create long, ordered trails.
No ant is really aware of the big picture. But when one pheromone sniff is multiplied 40,000 times, it becomes a powerful force. "The ability to do these collective tasks arises from very simple, layered behaviors," said David Bruemmer, a computer scientist on the project. It's kind of like The Wave at a football game.
So the next question for the researchers and engineers was, what individual behaviors in a robot will, collectively, lead to desired group behavior? In biology, it is almost always a simple response to environment—to a sound or a smell or another individual. Each ant really only knows what's ahead of and underneath it--the ant next to it and the piece of food two inches away. It responds only to its local environment.
The team of Dudenhoeffer, Bruemmer, Mark McKay, Matthew Anderson and Midge Davis have created a group of robots that work in the same way. These small creatures--each about as big as a chipmunk--roll on two big wheels and one small one and respond to light, water and sound.
The researchers take pride in how simple the creatures are. The robots are deliberately underdesigned. They wanted to see how simple they could make the robots and still end up with the desired collective behavior. If the robots will be searching for land mines or radiation spills, they need to be cheap. "Our goal is to make something that is literally disposable," said Bruemmer.
Plus, the simplicity of the robots allows them to stay small. The development team designed this group of mini-robots to find a contaminated wet spill on the floor. Often, the spills are in hard-to-reach areas. "Many of the environments that the Department of Energy faces are cluttered with piping," said Bruemmer. The wallet-sized rovers are small enough to maneuver through piping, but still big enough to travel in a reasonable amount of time.
On a quest for a spill, each mini-robot searches the area in a nearly random motion. When one rolls onto a wet spot on the floor, the water connects a circuit between its two front wheels and prompts the robot to stop moving and start chirping. When another robot hears chirping, it moves toward the sound. Eventually, the robots are arrayed outside the perimeter of the spill, chirping.
Hanging with the crowd
Sound simple? It's not. The robots need to stay within hearing distance of each other. And they also need to know not to run right into each other, a wall or a chair. To tackle the problem of staying close, but not too close, the scientists applied to the robots what they call social potential fields--one of their most innovative, and powerful tools for helping the robots mimic biological societies.
"Social potential fields are a combination of attraction and repulsion," said Dudenhoeffer, who pauses for a second and smiles, "like all social behaviors."
It's as if they have given each robot both a desire to hang out with the crowd and a desire for personal space--the same kind of behavior that birds demonstrate when flocking.
The robots keep their personal space free by sensing changes in light. When a creature--a moth, for example--approaches a wall, it senses the decrease in light intensity in front of it. The moth responds to that light change automatically, and steers out of the way. "It's a very intrinsic method of sensing," said Bruemmer. Many insects are born with the skill, and with the help of two small photoresistors, the mini-robots sense light in the same way.
In addition, just as living things adjust their sensitivity to light when in a dark or crowded room, the robots are continually fine-tuning how jumpy or brave they need to be. Otherwise, the robots would be turning when they didn't need to, or hitting obstacles all too often.
On the other hand, the group of robots stays close together through a sensitivity to sound. They chirp and tell each other to "come-hither." How often the robots turn also affects how close they stay to each other. The higher the turn frequency, the more clustered the robots remain. The lower the turn frequency, the more they tend to spread out.
By attributing these very simple insect-like behaviors to each robot, the robot designers have created a group of creatures that find a spill together, and in the process, neither jostle a neighbor nor stray too far from the pack. "They always succeed, eventually," said Bruemmer. "They always get there in a different way, but they always get there."
And the more robots that search together, the faster and more consistently they work. Where one robot will take two hours to eternity to find a spill, a group of nine robots will do it consistently in three to five-and-a-half minutes. As a group, they are more reliable, because as Bruemmer noted, "One of the robots may be having a bad day, but they're not all going to."
Not for every job
So far, the purely reactive behavior works well for finding a spill. Ultimately, the robots can be used to track a variety of chemicals. But the INEEL researchers know that their robots aren't the silver bullet for every robotic task. "We also get disadvantages, severe disadvantages," said Dudenhoeffer. These robots can't really give feedback.
"How do you query a colony of ants?" Dudenhoeffer asked. Getting feedback from the small, autonomous robots has been one of the engineers' greatest challenges. If a mini-robot on a monitoring mission finds something out of the ordinary, how does it let the human operator know? And how does the operator know if a robot falls down the stairs?
Bruemmer and Dudenhoeffer have come up with an intermediary that they hope mitigates some of the challenges. They use a larger, more intelligent, more expensive robot that can both deploy the smaller robots, lead them to a new area, and track their movements with a camera.
"We call him Junior," said Dudenhoeffer, "It's a sort of mother duck idea." The larger robot interacts with the human operator, and can both observe and lead (by chirping) the crowd of smaller, simpler robots.
It's still not like talking to each individual robot. But using Junior offers some control without overwhelming the operator. "We basically say, how can we get the best of both worlds–from fully reactive to very deliberative?" said Dudenhoeffer.
The best robot combination is obviously a group that makes the human's job easier, not harder. The robot army at INEEL is a joint collaboration between the scientists in robotics and in human factors—those who evaluate the human side of technology.
Human factors scientists are skilled at asking the questions that those in technology often forget to ask, such as, "How will humans respond to this technology?" "Does this technology really facilitate a person's work, or will it hinder it?" For example, in some cases, email communication helps people work faster. But email chain letters and email viruses often hinder the same work. Just because it is a technology with the potential to help doesn't mean that humans will always use it that way. Most technologies tread a fine line between how much work they require from humans, and how much they remove.
For the robots, Dudenhoeffer noted, "You don't want to take away what humans add. Some people insert a technology, but don't fully understand the human implications. We're not trying to create a robot that does everything, but a robotics system that allows both the human and robot to do what they do best."
Seed money for this project was provided by the INEEL Laboratory Directed Research and Development fund, and the Defense Advanced Research Programs Agency Software for Distributed Robotics.
The INEEL scientists have found that creating a group of small mini-robots is more effective if they think of it more as a swarm than an army. Each is then trained to work on its own according to certain rules and a collective behavior emerges from the masses of individuals acting independently.
The small INEEL puddle-finders are, so far as robots go, nothing special. But as a group, they represent a unique approach to helping humans with their most repetitive and dangerous jobs. Herds of mini-robots could be especially useful in environmental surveillance, said Dudenhoeffer, checking for gas leaks, radiation contamination or oil spills. They could eventually monitor demilitarized zones, or find wounded firemen in burning homes.
"We're still far away from the robustness of an ant colony," said Bruemmer, but the INEEL researchers have come a long way in creating a robot as independent, and disposable, as the lowly ant.