UCLA researchers have developed a new framework based on deep learning that enables a robot to tackle paper folding and the oriental art of origami.

Wherever we go, we come into contact with thin, flexible structures. Large deformations are a common feature of these structures when subjected to even relatively weak forces like gravity. Human beings have an astonishingly deep inherent awareness of the dynamics of such malleable objects. Getting robots to act with more human intuition remains an important area of ​​study, as it could lead to a wide range of useful applications for business and society.

It is not easy for robots to manipulate deformable objects because they need to predict how the object will change as it is manipulated in order to be successful. However, there are currently few solid answers for robotic manipulation of many other deformable things because most previous research has concentrated on cloth or string.

Recently, a group of researchers at the University of California, Los Angeles (UCLA) developed a novel computational framework that allows a robot to take on paper folding and the Asian art of origami.

Two of the most important investigations on this topic were previously carried out by research groups at Aalto University in Finland and the University of Bielefeld in Germany. The first study of his dealt with textiles, which are computationally easier to handle than paper. Rather, the paper is folded in seconds using a complex robotic system involving human-like manipulators.

The UCLA team was inspired to carry out this study due to the inadequacy of simple and efficient robotic paper folding systems. Consequently, the group set out to design a simple but potentially useful device that could fold paper using a single robotic manipulator.

The researchers present a robot control technique that teaches robots behaviors from a physical perspective, allowing them to take on jobs that require physically discerning manipulation with greater ease. And more specifically, they used offline environments to train artificial neural networks (ANNs) using physical simulations of paper folding. Throughout his training, the network became familiar with the “behavior” of a sheet of paper when held with various grips.

Training data was produced through mathematical and physical modeling on a computer. Subsequently, the trained neural network made rapid predictions online and in real time, leading to optimal manipulation trajectories. Scale analysis, borrowed from mathematics, is used to make non-dimensional predictions of the neural network, which is another novelty.

Non-dimensionalization is a mathematical physics technique that eliminates the need to worry about units of measure between input and output. There are no units for nondimensional quantity. Therefore, changing system drives will not affect the scan. Improves the generalization of the control framework, making it possible for the robot to fold sheets of paper with different thicknesses and geometries without additional training.

The “dimensionality” of the paper folding problem can be reduced by non-dimensionalization. In other words, it makes training easier while improving the real-time performance of the neural network.

An interesting result of this research is that physics-based scale analysis and machine learning algorithms work quite well together to manipulate deformable objects with robots. When it comes to paper, for example, the computational cost of using a traditional mathematical model of physics is intractable, making real-time manipulation impossible. However, suppose that machine learning is used without prior knowledge of the problem. In that case, a control scheme will be created that will only be effective for situations that match those of the training data.

According to the researchers, this framework is the first to use this synergistic approach. They hope that their study will be widely applied in various deformable manipulation tasks such as cable management, knot tying, robotic kirigami, etc. They plan to broaden their focus to include more advanced folding activities like robotic origami. Making it possible for a robot to fold paper into various shapes (paper planes, paper frogs, etc.) would be an intriguing endeavor.

review the Paper. All credit for this research goes to the researchers of this project. Also, don’t forget to join our 13k+ ML SubReddit, discord channel, Y electronic newsletterwhere we share the latest AI research news, exciting AI projects, and more.