Using simple robots to simulate genetic evolution over hundreds of generations, scientists shed light on why humans help each other.
Swiss scientists used tiny robots to simulate behavioral evolution over generations.
They have the first quantitative proof that altruistic behavior is determined by genetic proximity.
Their findings could lead to more responsive swarming robots.
For decades the prevailing wisdom in biology has been that individual self-sacrifice enables a group's genes to keep going in the long run. Now Swiss researchers have actually proven this is true by evolving generations of tiny autonomous robots that are confronted with a basic decision to share food.
"For biologists, this rule had never been tested empirically before in a quantitative way," said Laurent Keller, a biologist in the University of Lausanne's Department of Ecology and Evolution who worked on the study, published today in PLoS Biology. "It's the first field test to show it's working."
In 1964, British evolutionary biologist W.D. Hamilton published what later became known as Hamilton's rule of kin selection. He said that the evolution of altruistic behavior -- sacrificing oneself for the greater good -- is determined by genetic proximity. Meaning that giving food to your siblings will help the genes live on, even if you don't.
Since it was impossible to measure altruism's costs and benefits using animals, Keller worked with Dario Floreano, a robotics professor at the Ecole Polytechnique Fédérale de Lausanne. They used a fleet of specially-designed robots around one cubic inch in size, called "Alice."
Each microbot has two wheels, a linear camera, a rechargeable lithium-polymer battery, and infrared distance sensors to detect objects within an inch of the robot. Floreano said the bots can be differentially controlled to move backward, forward and rotate on the spot.
The researchers and a group of Ph.D. students set up an experimental arena with eight autonomous robots and eight food items. The robots were controlled using an architecture based on neurons. Upon depositing the food along a wall, the bots were given "fitness rewards" and a choice between keeping them or sharing them with the group.
A camera recorded the bots' actions. Then the researchers entered the robot behavior into a computer simulator to see what would happen over 500 generations of artificial evolution. They also transferred the resulting control system back to the physical robots to bring the testing full circle.
"We found that, in all cases, a transition occurred exactly as predicted," Keller said. "We showed that even with complicated genetics, the robots followed the behavior predicted by this rule."
Recently the Swiss government's civil protection department asked the duo to create flying robots that can set up a radio bridge for a search and rescue operation on the ground, replacing the need for trucks. The researchers applied the algorithms they'd developed to 10 flying bots equipped with antennas and transmitters.
"Evolution came up with a very effective policy -- a control system for these robots which eventually we translated to the physical robots," Floreano said. They recently conducted a successful test flight.
"In the future, we'll work on establishing connection among multiple rescuers on the ground, and rescuers who move around a lot."
Dan Palmer is a computer science professor at John Carrol University who builds programs for understanding swarm behavior. "Here we have a situation where biology and computational science are merging," he said of the researchers' research.
"You steal an idea from nature, apply it to some other problem you can't solve, and see if millions of years of engineering in an evolutionary sense can solve it for you."