AI evolution: Robots creating robots
A team of researchers from the University of Cambridge has taken an important step in the field of innovative artificial intelligence. They have created a robot who can autonomously create better robots.
The researchers invented a robot ‘mother’ to design, create, and test robot ‘children.’ The original robot can assess the abilities of its children and designs future generations by incorporating certain traits over others to further evolve the offspring’s potential.
In the experiment, the mother’s task was to build 10 children out of different plastic cubes–robot genes–with a small motor inside that enables movement. Each child could be composed of between one and five total plastic cubes. The mother robot judged each of its inventions on how far they moved from their start point during a set amount of time. The ones that moved the furthest remained unchanged for the next generation while the other slower moving robots were redesigned.
Over five separate, successive experiments, the team found that the average speed of the fastest child in the final generation was more than twice as fast as the average speed of the fastest child in the first.
The mother’s method of enhancing the abilities of successive generations is an AI replication of how species evolve in nature. Some genes are modified, added, deleted, or merged. The most successful offspring remain unchanged in order to preserve their abilities while mutation and crossover were introduced in the less successful.
“Natural selection is basically reproduction, assessment, reproduction, assessment and so on,” says lead researcher Dr. Fumiya Iida of Cambridge’s Department of Engineering, in a press release. “That’s essentially what this robot is doing–we can actually watch the improvement and diversification of the species.”
The study is published in PLOS One.
Source: AI evolution: Robots creating robots
Via: Google Alerts for AI