A robotic hand can manipulate complicated objects such as toy planes and Rubik鈥檚 cubes without dropping them, thanks to a sophisticated computer algorithm.
at the University of California, Berkeley, and his colleagues say their system, called RotateIt, works better than previous ones because it can rotate objects in three axes.
Advertisement
Qi began working on the robotic hand with the long-term goal that it could act as a home assistant for people. 鈥淚f we have a robot that has the same level of dexterity we have, and it can basically use anything we have, then it can better help us,鈥� he says.
The researchers trained the system in a simulation, where it was given details about object shapes and sizes. They then trained a neural network to try to rotate the object in the simulation as much as possible.
When the system was put into real-world practice, it was given visual and touch inputs, providing more fine-grained detail on where an object was. That enabled the hand to manipulate them more precisely.
鈥淭raditional grasping of an object takes place by calculating the centre of mass and having some broad idea about how the object is to be placed,鈥� says at the University of Sheffield, UK. 鈥淗owever, this is not always an easy problem to solve.鈥�
He says the method put forward by Qi鈥檚 team is different, enabling the speedy handling of unusual objects. 鈥淭his algorithm provides an interesting and easily accessible method for changing the grasp on an object once it has been picked up,鈥� says Aitken. 鈥淚t will see wide uses, especially on automated manufacturing lines, where it will prevent the need for components to be placed in an ordered manner.鈥�
Qi says more work is needed on thinner, slender objects such as pencils and screwdrivers, which the system can鈥檛 handle well. 鈥淭he skills we need to rotate them is different from the skills to rotate other objects.鈥�
arXiv
