“Evolving robots” teach us about extinct animals—and may end up an important battlefield technology.
Photograph by John Long.
When I tell people that I work on “evolving robots,” their common response is to joke—semi-seriously—about Skynet or some other sci-fi nightmare in which machines develop self-awareness and rebel against their human creators. I usually reply, in my best Borg-like patter, “Lower your shields and surrender your ships. Resistance is futile.”
But evolving robots are neither a joke nor just a Hollywood story device. My fellow researchers and I are using them to harness evolution, putting it to work as an automatic, hands-off process to go where no robot has gone before: the ancient past of animals and the unknown future of human technology.
Making robots that can evolve solves a serious problem that has long vexed biologists: Dead fossils tell no tales. While fossils inform us about evolutionary patterns, they don’t tell us about life’s processes, like the dynamics of physiology, behavior, and the “struggle for existence” that Darwin recognized as the basis of the evolutionary game of life. We can reconstruct and re-enact those missing processes using biorobots, a special class of physically embodied and fully autonomous machines designed to mimic living and behaving animals.
At first blush, this field, called evolutionary biorobotics, seems to present a Zen koan: How does one use evolution to study evolution? You can solve the riddle systematically, as I detail in my new book, Darwin’s Devices. First, you need a specific question about evolution. How did birds evolve flight? How did our vertebrate ancestors evolve from fish to legged land-lovers? Was the evolution of humanlike primates driven by selection for enhanced physical endurance during hunting?
Here’s one of my favorite questions: How did the first fishlike vertebrates evolve from wormlike ancestors some 500 million years ago? What we know comes from amazing soft-bodied fossils unearthed in the early Cambrian deposits of China. Schools of little inch-long fish, named Haikouichthys, had an internal support system, a fibrous head-to-tail rod called a notochord. The notochord is the scaffold upon which evolution builds the bony, jointed column of vertebrae that we know as our backbone. So to understand the origin of vertebrates, we have to figure out why bony vertebrae evolved in backbones.
To build a backbone for our biorobots, we’ve used a reverse-engineering technique called biomimetics. First, we studied extensively the structure and biomechanical function of the backbones of swimming sharks. Then, we replicated shark structure and function in artificial skeletons—biomimetic backbones made in the laboratory. Because we control the manufacturing process, we can create biomimetic backbones that are notochords lacking vertebrae, and backbones that have vertebrae.
If we are going to re-enact evolution, our biomimetic backbones need biorobots—in this case, ones that work like real fish. They have eyes and a special organ called a lateral line, which helps them sense predators in the water. Those sense organs are connected to a computerized brain that works just like the organic brain of fish, helping decide when to switch from feeding oneself to escaping from a predator. That switch in behavior requires that the biomimetic backbone, which is housed in the biorobot’s propulsive tail, alters its flapping to cruise, turn, or accelerate.
The biorobots, in turn, need an evolutionary arena to play the game of life. So we put our biorobots—many different kinds of evolving prey and an attacking predator—into a pool, turn them on, and let them go on their own. The winners are the prey that do the best job gathering food and, at the same time, avoiding being caught by the predators. Just as in evolution, winners get to produce more children than the losers. Nothing inappropriate going on here—the robotic reproduction happens on a computer, using a kind of software known as a genetic algorithm.
It turns out that biorobots rapidly evolve backbones with more vertebrae. Within five generations, the population of prey robots increases its average number of vertebrae. Because we can watch the biorobots behave in the water arena, we can see that more vertebrae allow them to swim and maneuver faster. Therefore, the evolution of the backbone in early vertebrates was likely driven by selection for enhanced feeding and fleeing behavior. Lost evolutionary process, revealed.
Mind-blowingly, the biorobots became more intelligent without evolving their brain. Though only their bodies evolved, they got better at feeding and fleeing. What gives? The lesson here is that brains are overrated. Intelligent behavior is created by the dynamic, ongoing interaction between the body, the brain, and the world in which the body-brain lives.
Photograph by John Long.
This process doesn’t just unmask mysteries of the past. It may also help create new kinds of robots that have intelligent behaviors, then tune those behaviors by allowing the body, the brain, or both to change in response to the evolutionary challenges presented. Just as we saw before, the trick is to ask the right question.
And eventually, evolving robots may even become important on the battlefield. The U.S. Department of Defense recently announced that it is looking for ways to build a “mobility platform” that can move from the surf zone to land and then through swamps and up steep, rocky terrain. This is the amphibious equivalent of asking for a flying car that can also wash your dishes. One way to solve this challenge may be to start with a known biological solution, like a sea turtle, and then evolve a novel design specifically for the performance requirements that extend beyond what turtles can do.
Furthermore, since evolutionary change comes in response to behavioral interactions, industrial engineers may begin to create robots that evolve their hardware as the conditions on the battlefield change. There are some significant hurdles to this, such as being able to quickly and automatically manufacture robots. Rapid prototyping via 3-D printing holds promise, but parts still need to be assembled by someone, and printers must be capable of working with a wide range of materials. This manufacturing bottleneck will initially constrain the battlefield evolution of robots to designs that are structurally simple. Some of the slack can be picked up by the co-evolution of the robots’ software programs, but, as we’ve seen, the neural control system is only part of what creates intelligent behavior. Since combining the evolution of brains and bodies will permit rapid and innovative behavioral adaptation during battle, it’s highly likely that defense programs will soon be putting evolution to work in designing their robotic weapons systems.
John Long is author of the new book Darwin’s Devices: What Evolving Robots Can Teach Us About the History of Life and the Future of Technology. He is director and co-founder of the Interdisciplinary Robotics Research Laboratory at Vassar College, where he also serves as the chair of the biology department and professor in the cognitive science program.