Evolutionary robotics is a field devoted to using lessons from evolution to develop autonomous robots that can produce more efficient robots. While that may sound like science fiction, it’s quite real, and at Cambridge University, a team has developed a robot that does just that.

Their robot acts as a “mother,” building “children” robots and then testing them, using the best traits she observes in that generation of robots and applying it to the next generation. In this case, the trait in question is speed: the children are tested on how far they move in a set amount of time. And over ten generations, the mother robot was able to arrive at an efficient design that, according to the researchers involved in the project, humans wouldn’t have come up with.

For this study, the robot built and studied 500 other robots, improving them with each generation. Each of those robots took about 10 minutes to design and build, so to speed things up, they plan on using computer models in future experiments. That doesn’t seem as fun, but it will allow evolutionary robotics to develop faster, which seems fitting.

As to applications, the team can already see a few of them. In manufacturing, for example, if robots can observe each other building cars, they can create new robots that make use of the best techniques and design from the current generation to do the job better in the future.

Of course, we use robots in a number of fields and industries, and as robotics improves over time, they’ll likely be showing up in more and more places. If those robots can improve themselves, it leaves humans with time to work on other things, like new robots. But it also means those robots that do dangerous jobs can get better and more efficient, or those that assist doctors can learn on the job.