The origin of the robot species

Robots "evolve" over 10 generations to perform a task twice as fast
August 12, 2015

A “mother robot” (A) is used for automatic assembly of candidate agents from active and passive modules. For the construction process, the robotic manipulator is equipped with a gripper and a glue supplier. Each agent is represented by the information stored in its genome (B). It contains one gene per module, and each gene contains information about the module types, construction parameters and motor control of the agent. A construction sequence encoded by one gene is shown in (C). First, the part of the robot which was encoded by the previous genes is rotated (C1 to C2). Second, the new module (here active) is picked from stock, rotated (C3), and eventually attached on top of the agent (C4). (credit: Luzius Brodbeck et al./PLOS ONE)

Researchers led by the University of Cambridge have built a mother robot that can build its own children, test which one does best, and automatically use the results to inform the design of the next generation — passing down preferential traits automatically.

Without any human intervention or computer simulation, beyond the initial command to build a robot capable of movement, the mother created children constructed of between one and five plastic cubes with a small motor inside.

In each of five separate experiments, the mother designed, built and tested generations of ten children, using the information gathered from one generation to inform the design of the next.

The results, reported in an open access paper in the journal PLOS One, found that the “fittest” individuals in the last generation performed a set task twice as quickly as the fittest individuals in the first generation.

Natural selection

Natural selection is ”essentially what this robot is doing — we can actually watch the improvement and diversification of the species,” said lead researcher Fumiya Iida of Cambridge’s Department of Engineering, who worked in collaboration with researchers at ETH Zurich.

For each robot child, there is a unique “genome” made up of a combination of between one and five different genes, which contains all of the information about the child’s shape, construction and motor commands.

As in nature, the evolution takes place through “mutation,” where components of one gene are modified or single genes are added or deleted, and “crossover,” where a new genome is formed by merging genes from two individuals.

To allow the mother to determine which children were the fittest, each child was tested on how far it traveled from its starting position in a given amount of time. The most successful individuals in each generation remained unchanged in the next generation to preserve their abilities, while mutation and crossover were introduced in the less successful children.

The increase in performance was due to both the fine-tuning of design parameters and the fact that the mother was able to invent new shapes and gait patterns for the children over time, including some designs that a human designer would not have been able to build.

Cambridge University | Fumiya Iida’s research looks at how robotics can be improved by taking inspiration from nature, whether that’s learning about intelligence, or finding ways to improve robotic locomotion. Iida’s lab is filled with a wide array of hopping robots, which may take their inspiration from grasshoppers, humans or even dinosaurs. One of his group’s developments, the “Chairless Chair,” is a wearable device that allows users to “sit” anywhere, without the need for a real chair.

Creative machines

“One of the big questions in biology is how intelligence came about — we’re using robotics to explore this mystery,” said Iida. “We think of robots as performing repetitive tasks, and they’re typically designed for mass production instead of mass customization, but we want to see robots that are capable of innovation and creativity.”

In nature, organisms are able to adapt their physical characteristics to their environment over time. These adaptations allow biological organisms to survive in a wide variety of different environments — allowing animals to make the move from living in the water to living on land, for instance.

But machines are not adaptable in the same way. They are essentially stuck in one shape for their entire “lives,” and it’s uncertain whether changing their shape would make them more adaptable to changing environments.

Using a computer simulation to study artificial evolution generates thousands, or even millions, of possibilities in a short amount of time, but the researchers found that having the robot generate its own possibilities, without any computer simulation, resulted in more successful children. The disadvantage is that it takes time: each child took the robot about 10 minutes to design, build and test. A robot also requires between ten and 100 times more energy than an animal to do the same thing.

According to Iida, in the future they might use a computer simulation to pre-select the most promising candidates, and use real-world models for actual testing.

Cambridge University | Researchers have observed the process of evolution by natural selection at work in robots, by constructing a “mother” robot that can design, build and test its own “children,” and then use the results to improve the performance of the next generation, without relying on computer simulation or human intervention.

 Abstract of Morphological Evolution of Physical Robots through Model-Free Phenotype Development

Artificial evolution of physical systems is a stochastic optimization method in which physical machines are iteratively adapted to a target function. The key for a meaningful design optimization is the capability to build variations of physical machines through the course of the evolutionary process. The optimization in turn no longer relies on complex physics models that are prone to the reality gap, a mismatch between simulated and real-world behavior. We report model-free development and evaluation of phenotypes in the artificial evolution of physical systems, in which a mother robot autonomously designs and assembles locomotion agents. The locomotion agents are automatically placed in the testing environment and their locomotion behavior is analyzed in the real world. This feedback is used for the design of the next iteration. Through experiments with a total of 500 autonomously built locomotion agents, this article shows diversification of morphology and behavior of physical robots for the improvement of functionality with limited resources.