This Three-Armed Robot Can Slice-and-Dice Salmon Into Sashimi

Researchers built Sashimi-Bot to master one of robotics’ slipperiest challenges.

A salmon loin usually slumps, sticks, slides and gives way under pressure, which is why even a clean slice of sashimi depends on a chef’s practiced touch.

Now researchers have taught a three-armed robot to do the job on its own. It can shape a slippery piece of fish, cut it with a chef’s knife, sense when the blade reaches the board, and lift each raw slice with chopsticks.

The system, called Sashimi-Bot, shows how robots may begin to handle the soft, fragile materials—food, crops, fabric and perhaps even tissue—where today’s machines still struggle.

Salmon Difficulty Level

Most robots do best with rigid, predictable objects. They can manipulate boxes, carts, and even screws. Salmon is a harder ask. It changes shape when pushed, slides under tools, and can be easily ruined by too much pressure.

Sashimi-Bot uses three robotic arms to divide the job. One arm holds tools, including a chef’s knife. A second steadies the salmon with a gentle stabilizer. A third uses chopsticks to pick up each slice and place it on a tray.

Before the robot made its first cut, it had to get the loin under control.

Sashimi-Bot used one arm to nudge the salmon into a centered, orderly shape on the cutting board, so the knife could cut at the right angle. That shaping skill was learned in simulation through deep reinforcement learning, a trial-and-error AI method in which the software improves by testing actions and receiving rewards for better ones. The team then transferred the system to the real robot without giving it extra practice on actual salmon.

A Sense of Touch

Cutting introduced another challenge: the robot needed to know when the knife had reached the cutting board. A soft gripper can hold a regular chef’s knife, but it also makes the knife’s exact position slightly uncertain.

×

Get smarter every day...

Stay ahead with ZME Science and subscribe.

Daily Newsletter
The science you need to know, every weekday.

Weekly Newsletter
A week in science, all in one place. Sends every Sunday.
No spam, ever. Unsubscribe anytime. Review our Privacy Policy.

Thank you! One more thing...

Please check your inbox and confirm your subscription.

The team added a GelSight tactile sensor, a device that uses a soft gel surface and a camera to detect pressure and deformation. In effect, it gave the knife hand a form of touch.

The researchers trained the sensor system on 12,397 readings from 157 cutting motions. 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%.

In the full test, Sashimi-Bot cut 34 salmon slices. The slices ranged from 6 to 16 millimeters thick, with the knife tilted up to 20°. Six slices stuck to the knife blade; the robot recovered all six by picking them directly from the blade. Of the 28 slices left on the board, it picked up 26 successfully. The two failures came from very thin slices slipping from the chopsticks.

RelatedPosts

The AI Boom Is Eating the Power Grid
AI Experts Predict Machines Could Outthink Humans by 2040. But Some Are Betting on Even Sooner
Artificial intelligence should be protected by human rights, Oxford mathematician argues
Google just let an Artificial Intelligence take care of cooling a data center

More Than a Sushi Stunt

The robot is not ready to replace a sushi chef. The setup moved slowly under research safety limits: a cut took about 5.3 seconds, but the full cycle between cuts averaged 27.9 seconds, or 37.7 seconds when a slice had to be retrieved from the knife.

Still, this proof of concept shows where things are heading in the industry. Your run-of-the-mill robot struggles with many soft materials. A system that can cut, stabilize, sense, and pick up deformable objects could eventually help automate food processing closer to production sites.

The study was published in the journal npj Robotics.