Researchers from the College of Southern California (USC) Division of Laptop Science and NVIDIA have unveiled a brand new simulator for robotic reducing that may precisely reproduce the forces appearing on a knife because it slices by frequent foodstuffs, reminiscent of fruit and greens. The system may additionally simulate reducing by human tissue, providing potential purposes in surgical robotics. The paper was introduced on the Robotics: Science and Programs (RSS) Convention 2021 on July 16, the place it obtained the Greatest Scholar Paper Award.
Up to now, researchers have had hassle creating clever robots that replicate reducing. One problem: in the true world, no two objects are the identical, and present robotic reducing techniques wrestle with variation. To beat this, the crew devised a novel method to simulate reducing by introducing springs between the 2 halves of the item being minimize, represented by a mesh. These springs are weakened over time in proportion to the pressure exerted by the knife on the mesh.
“What makes ours a particular type of simulator is that it’s ‘differentiable,’ which signifies that it could possibly assist us routinely tune these simulation parameters from real-world measurements,” mentioned lead creator Eric Heiden, a PhD in laptop science pupil at USC. “That is vital as a result of closing this actuality hole is a big problem for roboticists at this time. With out this, robots could by no means escape of simulation into the true world.”
To switch expertise from simulation to actuality, the simulator should have the ability to mannequin an actual system. In one of many experiments, the researchers used a dataset of pressure profiles from a bodily robotic to supply extremely correct predictions of how the knife would transfer in actual life. Along with purposes within the meals processing business, the place robots may take over harmful duties like repetitive reducing, the simulator may enhance pressure haptic suggestions accuracy in surgical robots, serving to to information surgeons and forestall harm.
“Right here, you will need to have an correct mannequin of the reducing course of and to have the ability to realistically reproduce the forces appearing on the reducing software as totally different sorts of tissue are being minimize,” mentioned Heiden. “With our method, we’re capable of routinely tune our simulator to match such several types of materials and obtain extremely correct simulations of the pressure profile.” In ongoing analysis, the crew is making use of the system to real-world robots.
Co-authors are Miles Macklin, Yashraj S Narang, Dieter Fox, Animesh Garg, Fabio Ramos, all of NVIDIA.
The total paper (open entry) and weblog put up can be found right here: https://diff-cutting-sim.github.io/