Full Breakdown
Sashimi-Bot Slices Salmon with Tactile Sensing
6/21/2026, 12:13:43 PM
Automated Sashimi Production
NTNU’s three-armed robot steadied a raw salmon loin, sliced it with a chef’s knife, and lifted each piece using chopsticks. In a single trial the system produced 34 slices (6–16 mm thick); 28 landed on the board, 26 were grasped, and six that stuck to the blade were retrieved directly.
Technical Approach and Results
The knife arm carried a GelSight tactile sensor—a soft gel surface with an embedded camera—that signals board contact. The sensor model was trained on 12,397 readings from 157 simulated cuts via deep reinforcement learning. On a held-out test set it achieved 95 % accuracy, 99 % precision and 67 % recall. Each cut took ~5.3 seconds; the full cut-and-pick cycle averaged 27.9 seconds, rising to 37.7 seconds when blade retrieval was needed. Overall success for board-landed slices was 93 % (26 of 28).
Official Statements & Responses
The research team reported that combining tactile feedback with reinforcement-learning control enabled high-precision slicing of a deformable object. They stressed that the system operates under strict safety limits and serves as a proof-of-concept rather than a production-ready tool. Published in *npj Robotics*, the work suggests broader use in food processing, agriculture, textiles and biomedical tissue handling.
Criticism & Opposition
Recall remained modest, leading to occasional missed board contacts. Cycle time is slow for industrial use, and very thin slices sometimes slipped from the chopsticks. The team acknowledged that the robot is not yet ready to replace a sushi chef, highlighting the gap between laboratory performance and commercial deployment.
Why It Matters
Demonstrating real-time tactile perception and adaptive motion on a soft, irregular material, Sashimi-Bot points to future automation of delicate tasks. Adoption could cut manual labor in food-processing lines, improve product consistency, and enable handling of other soft materials that have traditionally resisted robotic manipulation.
Verbatim Quotes
- “On a held-out test set, the model reached 95% accuracy in detecting board contact, with 99% precision, though its recall was lower at 67%.” — Research Team, Norwegian University of Science and Technology
- “It successfully grasped 26 of the 28 slices that fell onto the cutting board with chopsticks.” — Research Team, Norwegian University of Science and Technology
- “7 seconds when a slice had to be retrieved from the knife.” — Research Team, Norwegian University of Science and Technology
- “More Than a Sushi Stunt The robot is not ready to replace a sushi chef.” — Research Team, Norwegian University of Science and Technology
What’s Next
The authors aim to improve sensor recall, shorten cycle times and test the approach on other soft materials. Ongoing work will refine simulation-to-real-world transfer learning to broaden industrial applicability.
