Robots that learn to crawl before they walk

A roboticist from the University of Vermont has created an evolving robot. While studying how tadpoles change their bodies to become frogs in order to walk, he realized that the same principles can be applied to robots and he went on to design a robot simulation where he set loose a bunch of synthetic beasts that move around in a 3D space. He then programmed the robots with a genetic algorithm that experiments with different motions to reach an object without tipping over. The algorithm makes the robots experiment with different motions until it develops an efficient way of travelling to the light source. The different robots slithered, shuffled or walked up right – depending on their body plan in order to reach the light source.

At the end of the experiment, he realized that the robots that started out crawling before they walked or ran performed better than robots that began upright. They could even deal with unknown challenges, such as attempts to trip them over. With the results of the successful simulation, he decided to take it to the real world and build a simple robot using Lego Mindstorm kits and gave it an adaptive artificial intelligence. And what he has done is pretty impressive. Can you imagine robots children or pets that we can raise in the future? Check out the video demonstration of the robot below.

George Wong
Categories: Robots
Tags: experiment, Robots