Robots Learn to Share, Validating Hamilton's Rule
04.05.11 - An evolutionary robotics experiment at EPFL and UNIL supports Hamilton’s rule of altruism and improves swarm robotics collaboration as a result.
Using simple robots to simulate genetic evolution over hundreds of generations, researchers shed light on one of the most enduring puzzles in biology: Why do most social animals, including humans, go out of their way to help each other? Reporting in the May 3rd issue of the Public Library of Science (PLoS) Biology, EPFL robotics professor Dario Floreano teams up with University of Lausanne biologist Laurent Keller to weigh in on the oft-debated question of the evolution of altruism genes.
Altruism, the sacrificing of individual gains for the greater good, appears at first glance to go against the notion of “survival of the fittest.” But altruistic gene expres-sion is found in nature and is passed on from one generation to the next. Worker ants, for example, are sterile and make the ultimate altruistic sacrifice by not trans-mitting their genes at all in order to insure the survival of the queen’s genetic makeup.
Testing the evolution of altruism using quantitative studies in live organisms has been largely impossible because experiments need to span hundreds of generations and there are too many variables. However, Floreano’s robots evolve rapidly using simulated gene and genome functions and allow scientists to measure the costs and benefits associated with the trait.
In 1964, biologist W.D. Hamilton proposed a precise set of conditions under which altruistic behavior may evolve, now known as Hamilton’s rule of kin selection. Here’s the gist: If an individual family member shares food with the rest of the family, it reduces his or her personal likelihood of survival but increases the chances of family members passing on their genes, many of which are common to the entire family. Hamilton’s rule simply states that whether or not an organism shares its food with another depends on its genetic closeness (how many genes it shares) with the other organism.
“We have shown that Hamilton’s kin selection theory always accurately predicts the relationship between the evolution of altruism and the relatedness of individuals in a species,” explains Markus Waibel, lead author of the paper and former doctoral student of both Keller and Floreano.
Hamilton’s rule has long been a subject of much debate because its equation seems too simple to be true. “This study mirrors Hamilton’s rule remarkably well to ex-plain when an altruistic gene is passed on from one generation to the next, and when it is not,” says Keller.
Previous experiments by Waibel, Floreano and Keller showed that foraging robots doing simple tasks, such as pushing seed-like objects across the floor to a destina-tion, evolve over multiple generations. “These robots start out with a completely random series of coded information, the robot version of a genome, and only the code that works makes it to the next level,” explains Floreano. Those robots not able to push the seeds to the correct location are selected out and cannot pass on their code. Robots that perform comparatively better see their code reproduced, mutated, and recombined with that of other robots into the next generation—a minimal model of natural selection. After hundreds of generations, these robots become more efficient and even learn to work in groups without any further interference from the researchers.
The new study by EPFL and UNIL researchers adds a novel dimension: once a foraging robot pushes a seed to the proper destination, it can decide whether it wants to share it or not. Evolutionary experiments lasting 500 generations were repeated for several scenarios of altruistic interaction—how much is shared and to what cost for the individual—and of genetic relatedness in the population. The researchers created groups of relatedness that, in the robot world, would be the equivalent of complete clones, siblings, cousins and non-relatives. The groups that shared along the lines of Hamilton’s rule foraged better and passed their code onto the next generation.
The findings are already proving useful in swarm robotics. “We have been able to take this experiment and extract an algorithm that we can use to evolve cooperation in any type of robot,” explains Floreano. “We are using this altruism algorithm to improve the control system of our flying robots and we see that it allows them to effectively collaborate and fly in swarm formation more successfully.”
Photo Credit: EPFL/ Alain HerzogReturn to previous page
Rolex Learning Center
Case postale 122
1015 Lausanne 15
Tel: +41 (0)21 693 24 91