A restaurant's robot started missing grabs. Nobody could tell if it was a software bug or a broken part, until the data made the answer obvious.
That diagnostic story sits at the center of a detailed new writeup from Dyna Robotics. The company deploys AI robots doing repetitive tasks like napkin folding at chains including Din Tai Fung.
The Napkin-Folding Problem That Wasn't About Folding
Dyna says its newest model folds napkins 2.7 times faster than its previous version, at higher quality. But folding alone wasn't the hard part.
Folded napkins also need to land in the correct bin position, often one of ten nearly identical stacks. That requires genuine instruction-following, not just a folding motion.
The company could have trained a separate model for each restaurant's specific bin layout. Instead, it built general instruction-following into the model itself.
A Terabyte of Data a Day
Dyna's deployed fleet generates more than a terabyte of raw data daily. That includes camera streams, robot state, control commands and hardware telemetry from every shift.
The company runs a continuous, fixed-size recording window rather than only capturing formal task runs. Unexpected events between tasks still get captured this way.
Solving a Mystery With Automated Labels
A Din Tai Fung site's throughput started slipping recently, with nothing unusual showing up in standard logs. Dyna's automated labeling system, which segments every robot run into individual procedure steps, narrowed the problem down quickly.
Failures were climbing specifically in the step where the robot pulls a napkin off the stack. Nearly all of them were the same failure mode: missed grabs.
The cause turned out to be a worn gripper, not a software problem. Dyna was direct about the limits of its own tooling here.
The labeling system could locate exactly where the problem was happening. It couldn't tell the difference between a worn gripper and a flawed model policy on its own.
Miami's own restaurant robotics operators face that same diagnostic gap. Kitchen Robotics, a Miami startup behind the Beastro robotic kitchen platform, raised roughly $11.6 million before Munich-based Circus Group acquired it this April. Its automated dish prep runs in foodservice venues around the clock.
Better Days Robotic Kitchens is what Refresh Miami has called the city's first fully robot-run restaurant. Its cooking robots run 24 hours a day, preparing Mediterranean-inspired meals for local delivery apps. At that duty cycle, telling a worn part from a bad model update is not a hypothetical problem.
What Miami Businesses Should Ask Vendors
This is a genuinely useful look at what production AI robotics actually requires beyond the headline capability. Miami's growing base of restaurant, hospitality and logistics businesses evaluating robotics vendors should ask exactly these kinds of operational questions.
Can the vendor tell a hardware problem from a software one without guessing? That distinction is the difference between a quick fix and a fleet-wide mystery.
Nvidia Is Quietly Becoming AI's Biggest Financier