Original URL: http://www.theregister.co.uk/2014/05/13/robo_arm_catch/

Now that's PROPER SCIENCE: Boffins teach robo-arm to catch flying beer bottle

Learns how to calculate trajectories in real time

By Brid-Aine Parnell

Posted in Science, 13th May 2014 13:14 GMT

Video Youtube Video

A new robo-arm capable of catching anything that’s thrown at it – even at ultrafast speeds – has been unveiled by boffins from the École polytechnique fédérale de Lausanne (EPFL).

The super-fast grasping hand can react on the spot, snatching items like a tennis racket, a ball or half a bottle of beer out of the air in less than five-hundredths of a second.

The robo-arm, with its three joints and sophisticated four-fingered hand, waits with its palm open. Toss something over and it suddenly unwinds, working out complex shapes and trajectories to pluck the item out of the air.

"Increasingly present in our daily lives and used to perform various tasks, robots will be able to either catch or dodge complex objects in full-motion,” said Aude Billard, head of the learning algorithms and systems lab (LASA) at EPFL.

“Not only do we need machines able to react on the spot, but also to predict the moving object’s dynamics and generate a movement in the opposite direction."

Human beings react to thrown items instinctively, performing all the calculations necessary to catch a ball unconsciously*. But the ability to catch something requires a robot to take into consideration a number of different parameters and unforeseen events - and do it quickly.

"Today’s machines are often pre-programmed and cannot quickly assimilate data changes,” said Billard. “Consequently, their only choice is to recalculate the trajectories, which requires too much time from them in situations in which every fraction of a second can be decisive."

To speed the whole thing up, LASA researchers programmed the robo-arm to learn as humans do, by trial and error. Instead of giving the arm specific instructions for each catch, the boffins “showed” it possible trajectories then manually guided the arm to the target.

The exercises were repeated multiple times with five common objects - a ball, an empty bottle, a half-full bottle, a hammer and a tennis racket - representing a range of complex situations. In the case of the racket, the place you catch it - by the handle - is not the same as its centre of gravity, whereas with the half-full bottle, its centre of gravity changes over its trajectory.

The researchers lobbed the items at the robo-arm and the robot used a series of cameras around it to make a model for how to catch it, using trajectories, speed and rotational movement. That model is translated by the scientists into an equation the arm can use to position itself quickly whenever that object is thrown.

In the milliseconds the robo-arm has to catch the object, it refines and corrects the trajectory using real-time images and its basic learned technique for that item.

®

* YRMV