The evolving “Brain” for robots

Decades, people tried to copy the human mind, using increasingly clever software, creating an ever more sophisticated artificial intelligence system. But, despite all attempts, even the best artificial intelligence system slightly exceeds the average intelligence cockroach.

What’s the problem? Until now, all artificial intelligence systems based on software that worked on the constant hardware. All changes to existing hardware have been done through software. But a group of developers from the University of Oslo (Oslo University) in Norway has made a step, which may form the basis of the next generation hardware.

Professor Jim Torres (Jim Torresen) and Kierre Glut (Kyrre Gillette) built a developing hardware (Evolvable Hardware – EHW), Which use genetic algorithms. In other words, they created a computer hardware capable of has evolved.

The system of Torres and Getty genetic resources is that the hardware design, which may be the most effective in achieving the goal. In the real world may need 20-30 thousand generations before the biological system will find the perfect configuration for solving a problem, and it can take 8-9 hundred thousand years. The system created by the Norwegian team spends the same number of generations, only a few seconds.

The work on EHW began in the late 90-ies. With Torres started using evolutionary algorithms in 2004, when they made a robot-chicken “Henrietta”. The robot uses evolutionary principles in its software in order to learn how to do this or that action, without attempting to understand the world and create a solution through the use of artificial intelligence.

The robot “Henrietta” tried to apply arbitrary changes in their actions, correlating with the problem, and chose the best of them. Thus, following the best discoveries, he was able to solve their specific problems. Thanks to “Henrietta” able to understand better how it works evolution. Similarly, the hardware evolution system works to find a configuration that would be most effective in solving the problem.

Evolution can solve many problems that a programmer typically has to decide or provide opportunities to address them. For example, the robot was sent to Mars and fell into the pit. Using evolutionary methods, the robot could learn how to climb out of the pit without human assistance.

Now the group wants to develop a robot to assist in the installation of oil pipes and other oilfield equipment at a depth of 2.000 meters. The depth makes it almost impossible to direct contact with the robot. He should be 2-3 miles of wires stretching out behind him, or to transmit signals through echo, which in turn will give a considerable delay between the command and its execution. Evolutionary robot could find a solution to problems on site within a few seconds without operator intervention.