 |  Dexter learns by analysing 20,000 actions a second  |
Robot in action Dexter took its first tentative steps only a few days after it first discovered how to stand upright.
Dexter's designers say their robot differs from commercially available predecessors because it can learn from its mistakes.
It is the culmination of six years' work by Anybots, an independent research group of three engineers.
Founder, Trevor Blackwell, said: "When we started out Dexter had a very general idea of what a walking motion should look like.
"The first time it [tried] it just fell over right away.
"100 times a second we record about 200 different things: the position of the joints, the forces on the feet, and also the equivalent of what the inner ear measures: the way the body is tilting."
Jobs humans do
Dexter then analyses this information to modify its movements.
Dr Blackwell said walking robots currently on the market, such as Honda's Asimo, differ because their creators programmed their movements before they were switched on.
He said he was talking to industrial companies to develop Dexter for jobs people usually carry out in protective clothing.
The aim is to design a robot that can adapt to several environments and roles, like a human does, rather than requiring specific programming.
Before Dexter is ready for work it has to develop in a similar way to a child - with some coaxing but plenty of self-motivation.
'Running within months'
The developers are setting it new tasks all the time, said Dr Blackwell.
"We're trying to work like trainers do," he explained. "We think no, no, you've got to do that faster."
Anybots hopes Dexter will have taught itself to run within a few months.
If it does it will be following in Asimo's footsteps. The Japanese cousin mastered the art of speed more than two years ago and can run at a three kilometres an hour.
No comments:
Post a Comment