Artificial evolution will be used to design future robots, with machines choosing their own potential mates and producing offspring through rapid prototyping technologies.
The technique, under development at Bath University, is expected to be faster and more efficient than conventional human methods of designing robot hardware and software. So far ‘genetic’ algorithms have been used to create software, but these operate within certain parameters to guide development.
Now Dr Dylan Evans, leader of the Bath project, wants to use the principles of natural selection to create whole new machines.
‘In nature selection is a potential engine for generating new solutions for design problems.’
The work is funded by a £40,000 grant from the Engineering and Physical Sciences Research Council. Similar research has been conducted at other universities with so-called ‘predator’ and ‘prey’ robots, but Evans said this is only likely to develop insect-level machine intelligence.
‘To get higher life forms you need more complex emotions. We expect this new process to eventually design whole new control architectures.’
In this case complex ’emotions’ mean artificial sexual attraction and reproduction, where machines select a potential mate based on certain traits and characteristics. Previously artificial evolution just used random mutation.
Evans has developed the first few ’emotional’ brains using neural network software. His experiment involves robots ‘dancing’ to ‘attract’ a mate. Once two robots have decided they are compatible they will touch for five seconds to transfer their digital ‘DNA’.
The new software entity is then uploaded to a computer. Evans hopes that with every generation the dancing could become more sophisticated, creating more complex control codes.
The next step is to link the robots to a rapid prototyping machine, which will remove humans from the design loop.
For this Evans has applied for a further EPSRC grant of £200,000.