Researchers at the Department of Computer Science at the University of Southern California (USC) and NVIDIA have unveiled a new simulator for robotic cutting that can accurately reproduce the forces acting on a knife as it slices through ordinary foods such as fruits and vegetables. The system could also simulate human tissue excision, providing potential applications for surgical robotics. The paper was presented at the Robotics: Science and Systems (RSS) conference on July 16, 2021, where it received the Best Student Award.
In the past, researchers have had difficulty creating intelligent robots that repeat cutting. One challenge: no two objects are alike in the real world, and current robotic-cutting systems struggle with variation. To overcome this, the team developed a unique approach to simulating shear by bringing springs between two halves of a sheared object represented by a grid. These springs weaken over time with respect to the force exerted by the knife net.
“What makes our special simulator is that it is‘ separable ’, which means it can help us tune these simulation parameters from real-world measurements,” said Eric Heiden, a PhD student in computer science at USC. . “It’s important because bridging this reality gap is a major challenge for robotics today. Without this, robots will never be able to break out of simulation into real life.”
To transfer skills from simulation to reality, the simulator must be able to model a real system. In one experiment, researchers used a set of force profiles from a physical robot to produce highly accurate predictions of how a knife would move in real life. In addition to food industry applications where robots could take over hazardous tasks such as repetitive surgery, the simulator could improve the accuracy of force haptic feedback in surgical robots, help guide surgeons, and prevent injuries.
“Here, it is important to get an accurate model of the cutting process and to be able to realistically reproduce the forces acting on the cutting tool during the cutting of different tissues,” Heiden said. “With our approach, we are able to tune our simulator to automatically match such different types of materials and achieve very accurate force profile simulations.” In ongoing research, the team applies the system to real robots.
The authors are Miles Macklin, Yashraj S Narang, Dieter Fox, Animesh Garg, Fabio Ramos, all NVIDIA.
The full paper (open access) and blog post are available here: https://diff-cutting-sim.github.io/