DiSECt, introduced at the 2021 Robotics: Science and Systems (RSS) conference, can precisely cut fruits and vegetables with its algorithms. The robotic researchers of NVIDIA and the University of Southern California presented DiSECt as the first differentiable simulator for robotic cutting. The simulator predicts the pressure on the knife precisely when pressing and cutting natural soft items such as fruit and vegetables.

Recently, a group of researchers from USC Viterbi and MIT created a robotic assistant that can fix IKEA furniture and a robotic arm that helps us get dressed, respectively.

As it turns out, the cutting process requires adjustment to the stiffness of the items, the force exerted during cutting, and often a chopping movement to be cut. To do this, researchers employ a series of strategies that draw feedback to direct the evolution of the controller. However, for each instance of the same problem, fluid controller adaptation requires careful parameter tweaking. Though these strategies are successful in industrial environments, no two cucumbers (or tomatoes) are the same, making this family of algorithms worthless in a more generic context.

On the contrary, recent research focuses on developing differentiable control problems algorithms, which means that output sensitivity may be estimated without excessive sampling. When the simulated dynamics are distinguishable, but the technique of simulated cutting is not distinguishable, an efficient solution for control problems is attainable!

Differentiated cutting simulation is a concern as cutting is naturally a discontinuous process in which the creation of cracks and the propagation of fractures prevent gradation. This challenge is solved by offering a new way of continuously simulating crack development and mechanical damage.

The DiSECt simulator enhances the Finite Element Method (FEM) with a permanent contact model based on a signed distance field (SDF) and an ongoing damage model that inserts fountains over the opposite end to enable crack formation weakening to zero radiation.

Eric Heiden, a Ph.D. candidate at the Robotic Embedded Systems Lab at USC and an NVIDIA intern, says that

“In our ongoing research, we are bringing our differentiable simulation approach to real-world robotic cutting. We investigate a closed-loop control system where the simulator is updated online from force measurements while the robot is cutting foodstuffs. Through model-predictive planning and optimal control, we aim to find time-and energy-efficient cutting actions that apply to the physical system.”

The research was nominated for the Best Paper Award at RSS 2021. NVIDIA is an enterprise built on groundbreaking research and intelligent thinking. The focus is on AI, high-performance computers, self-driving cars, graphics, VRs, and increased reality with more than 200 scientists worldwide.