Industrial and household robots need to be able to manipulate all kinds of flexible and uneven objects. MIT researchers have come up with a system that enables robots to learn how to shape elasto-plastic materials like Play-Doh. It enables this by representing the object as a cloud of particles.
A graph-based neural network is used to learn and simulate the object’s dynamics. Robots can learn how to model shapes with 10 minutes of training data.
[HT]