A robot has learned how to mould modelling clay into letters that it has never seen before. Creating complex shapes out of doughy materials is a skill that could be put to use in the future in the form of a dumpling-making robot chef.
鈥淒eformable objects are ubiquitous in our daily life,鈥� says at the Massachusetts Institute of Technology. Robots capable of gently handling such objects could one day cook, do housework or even help care for elderly people, he says.
Advertisement
Li and his colleagues built a robot that could train itself to use two metal grippers to mould the approximate shapes of the letters B, R, T, X and A out of Play-Doh. The training involved just 10 minutes of randomly manipulating a block of the modelling clay beforehand, without requiring any human demonstrations.
While the robot used its metal grippers to randomly grab and squeeze the Play-Doh 150 times, four video cameras captured the interactions. The robot鈥檚 artificial intelligence used deep learning to convert the video data into a model of the clay block as a group of smaller particles and to predict how the lump of clay might change during future interactions.
The process allowed the robot to quickly learn about the clay鈥檚 characteristics and to plan for how it might create new shapes. It also avoided having to rely on more computationally intensive physics simulators.
The next big step for the robot involves learning how to make dumplings using dough and pre-made filling.
Forming dumplings poses new challenges, says Li. For example, the robot鈥檚 AI system must learn how to interact with many materials with different physical properties. The robotic body will also require upgrades beyond the two grippers to properly prepare dumplings.
Reference:
