A robotic arm with high-speed reflexes can catch almost anything thrown at it -- a ball, a tennis racquet, a bottle-- in just five-hundredths of a second, its Swiss developers say.

Researchers at the École Polytechnique Fédérale de Lausanne programmed the arm as a test of robotic capabilities to capture a moving object.

With quick reflexes, such robots could be valuable in factories, helping save human workers from accidents caused by flying objects, or they may be a key to self-driving vehicles where they could take control and avoid accidents, the researchers said.

"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," says Aude Billard, who heads the EPFL's Learning Algorithms and Systems Laboratory.

The robot arm -- about 4 feet long with three joints and a four-fingered hand -- can learn to catch things in the same way humans do, the researchers said, by trial and error.

First, the arm is manually guided to the projected target by a human, and the exercise is repeated several times to give the robot experience in the possible trajectories that will capture the target object.

Surrounding cameras capture the action, providing data that allows the robot to create a model of the thrown objects' movements based on their speed and trajectory.

The camera tracking system is used to refine the robot's actions, and the thrown object is snatched before it can hit the ground.

A variety of objects including the tennis racquet, a half-full bottle, a hammer and a ball were chosen to give the robot a challenge since for many of the objects thrown, the place it must be caught -- such as the racquet's handle -- does not correspond to its center of gravity.

That makes for complex movements in the air often involving rotation about several axes, the researchers said, requiring quick response not always present in current robot designs.

"Today's machines are often pre-programmed and cannot quickly assimilate data changes," Billard says. "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."

The research has been reported in the journal IEEE Transactions on Robotics.