Virtual Limbed Creatures Key to Legged Robots
CREDIT: Robot boy and dog via Shutterstock
Legged vehicles that clamber up mountains and possess the instincts of animals could someday be possible thanks to the first computer simulations of organisms that rely on articulated limbs to move.
Nicolas Chaumont , a graduate student at the Keck Institute in California, isn't trying to reinvent the wheel so much as replace it. Vehicles with legs could go where wheeled vehicles cannot – which could be vital for tasks such as rescue work, Chaumont says, because "70 percent of the land mass on Earth is completely unreachable by wheeled vehicles."
They are hoping their computer simulations will show the way to design this new generation of locomotion.
"It is so hard to design something that can go forward with legs reliably that we still don’t have any type of transportation that uses legs," Chaumont said. "With legs you have to coordinate the limbs so they can drive the robot forward in a very stringent manner. With wheels we just have to change the rate at which one wheel is turning."
Chaumont and his colleagues have created a virtual organism resembling a snake that crawls on the floor of a 3D environment towards a balled, green energy source. By changing the environment to make it more challenging for the virtual creature, the team hopes it will in time “evolve” legs that allow it to more easily navigate the simulated terrain. Legged virtual bots in turn could serve as blueprints for real-world biped, quadruped and spider-shaped robots that scientists have striven so long to create.
Wheeled vehicles, including tanks with tracks, can move only in straight lines and often fail upon reaching stairwells. Legged robots, by contrast, theoretically could climb rugged terrain to help fight forest fires or aid in rescue missions in areas inaccessible to cars and even helicopters.
"If this technology of legged locomotion someday becomes available, it is going to be very interesting," Chaumont said.
Instead of designing robots using established principles in engineering or anatomy, Chaumont and others in the field create virtual organisms that roam free in virtual environments where they must adapt to survive.
"For any modification made to the environment, evolution is going to push the population to adapt to these changes," Chaumont said. "If we want robots to acquire a set of skills, we don't need to redesign them, we modify the environment so that the desired skills will be selected for, and evolution is going to take care of the rest."
Once the desired behavior thrives in the simulation, scientists can perform real-world experiments with robots whose "brains" have been uploaded and body parts modeled from the system's winning populations. In this way, Chaumont and colleagues hope to discover nuances in neural and structural coordination that Nature took billions of years to evolve.
As simulations improve, "cognitive" skills such as common sense will develop in those populations that avoided dying thousands of times in treacherous virtual environments, Chaumont said.
Such skills will be vital for legged vehicles because unlike cars, a twenty-foot tall robot that can step over snow banks and ravines will have to be able to gauge terrain on its own.
"You will not be able to drive the robot against a wall like with a car. It will look much more like a horse. It will have instinct," Chaumont said. "And you would ride the robots like you ride a horse."
The research has been submitted to the journal IEEE Transactions on Evolutionary Computation.