Robotics researchers from NVIDIA and the University of Southern California (USC) announced today the first differentiable simulator for robotic cutting, or DiSECt for short. This new simulator can predict forces that will act on the knife as it pushes and slices through soft materials like fruits and vegetables.
Your first reaction might be, why do they need all that simulator science when you can just install a sharp blade on a robotic arm and smash it down? That’s certainly one solution, but part of the reason robot researchers like NVIDIA, and Sony and Panasonic all work with food is because food is oddly-shaped, has different textures and is delicate. If a robot can successfully work with soft objects like food, it can carry those techniques over to other applications like surgery (where plunging knives down is frowned upon).
Cutting through food with precision and care is actually quite complex. It requires feedback, adaptation, motion control and parameter setting as the knife makes its way through the object. Additionally, since each piece of fruit or vegetable is unique, the robot needs to adjust its cutting with each new object.
NVIDIA shared with us an advanced look at an article explaining the DiSECt research that was recently presented at 2021 Robotics: Science and Systems (RSS) conference. I’m not going to lie, it is dense and jargon heavy with paragraphs like this:
DiSECt implements the commonly used Finite Element Method (FEM) to simulate deformable materials, such as foodstuffs. The object to be cut is represented by a 3D mesh which consists of tetrahedral elements. Along the cutting surface we slice the mesh following the Virtual Node Algorithm [4]. This algorithm duplicates the mesh elements that intersect the cutting surface, and adds additional, so-called “virtual” vertices on the edges where these elements are cut. The virtual nodes add extra degrees of freedom to accurately simulate the contact dynamics of the knife when it presses and slices through the mesh.
But rather than focusing on the specifics of the research, there are some broader takeaways anyone in food tech can appreciate. First, DiSECt illustrates the continued importance of simulation and synthetic data in training robots. NVIDIA has actually built a kitchen as a training ground for its robots where it uses synthetic data and computerized simulation to virtually teach a robot tasks like identifying and picking up a box of Cheeze-Its. Similarly, DiSECt trains a robot to use a knife through simulation first, which can then be applied to the cutting object in the real world.
Additionally, giving robots added abilities will make them more useful in taking over dangerous tasks like repetitive cutting. Right now, robots in restaurants are frying foods and even making pizzas, but they aren’t doing more highly skilled, precision tasks such as cutting and slicing. A robot can’t get injured while cutting and could bring more safety to restaurant kitchens.
The good news for those interested in this type of cutting-edge research is that NVIDIA and USC are not the only companies doing work in this field. In 2019, researchers from Iowa State University published a similar paper on the intricacies of robot slicing.
Leave a Reply