Helping computers understand our 3D world

  • Date

    Mon 17 Feb 20

A thick piece of twisted brown rope tied in to a large knot.

Learning how to untangle knots could be the next step towards computers having a better understanding of our 3D world, according to a new research project at the University of Essex.

Dr Alexei Vernitski, from the Department of Mathematical Sciences, have been awarded a £196,000 grant from the Leverhulme Trust to use advances in machine learning to investigate how we can train computers so they can unravel knots on their own without human involvement.

Computers having a better understanding of our 3D environment could have many useful applications in the future.

Although humans have an excellent understanding of the 3D world, we are very bad at untangling. However, when presented with 2D pictures of knots, computers can untangle knots much better than humans – thanks to humans developing computer algorithms specifically for untangling knots.

Working with a computer scientist Dr Alexei Lisitsa, from the University of Liverpool, Dr Vernitski will be developing artificial intelligence to train computers to recognise rotated, twisted and tangled objects and then learn how to untangle them.

Currently, the most successful technology for computer vision is deep neural networks, which will be used in this three-year project so computers learn to accentuate the parts of the rope that should be pulled to untangle the knot.

Dr Vernitski explained: “This research is inspired by recent achievements of deep neural networks. Now, computers only need to know the rules of challenging games, such as chess, to be able to train themselves, without humans teaching them. Likewise, we will give computers a few basic facts about the 3D space, and see how computers will develop their knowledge.”