Building robots that survive the warehouse

Engineering & simulationNo Magic VLA (Vision Language Action models)0 churn within deployed systems

No Magic uses VLA models as a secondary recovery workflow to allow robots to autonomously solve packing errors in real-time by interpreting visual input and generating action plans, reducing the need for human intervention.

“we have, uh, censoring technology that's, you know, with AI that's constantly looking for these types of scenarios and, you know, flagging the system to switch to VLA mode, essentially. And then the robot can drive down and solve with that input…”

Engineering & simulationNo Magic AI models trained on production dataover a year in production

No Magic trains its physical AI models using real-world production data from deployed systems rather than just simulation, allowing the robots to generalize learnings and handle edge cases like specific shoebox orientations without retraining.

“we've built up a more general model of, of what uh, different types of products are and classifications of products based off of what we've seen in the real world with our uh, production data. We start there. So we can immediately begin picking when we come on site.”

Quality & testNo Magic Spot (vision sensor)

No Magic deploys its Spot vision sensor between pick and pack lines to perform order item counting and verification, identifying potential errors before shipping to reduce error rates.

“spot is just a vision sensor that you can put in between your pick and your pack lines to be able to do order item counting, order verification. Um, It's something that, uh, allows you to have a lot more data about what's going on in your, in your building…”