NTNU's Sashimi-Bot Cuts 34 Salmon Slices With 95% Tactile Accuracy
Updated
Updated · Interesting Engineering · Jun 19
NTNU's Sashimi-Bot Cuts 34 Salmon Slices With 95% Tactile Accuracy
3 articles · Updated · Interesting Engineering · Jun 19
Summary
Norwegian University of Science and Technology researchers tested Sashimi-Bot on real salmon loins, producing 34 sashimi slices and transferring 26 of 28 board-held pieces to a tray with minimal human help.
Three robotic arms handle positioning, knife work and plating, while a GelSight tactile sensor lets the system detect when the blade reaches the cutting board and stop before cutting too deeply.
More than 12,000 training samples from 157 cutting motions produced a contact-detection model with 95% accuracy and 99% precision, after the robot first learned fish positioning in simulation through deep reinforcement learning.
Six slices that stuck to the knife were all recovered directly from the blade; the only two transfer failures were extremely thin pieces that slipped from the chopsticks.
Published in npj Robotics, the project targets a broader robotics challenge—handling soft, deformable materials—with potential uses beyond food processing, including healthcare.