Inside the state of warehouse automation

Engineering & simulationCovariant Foundation Model

Covariant uses a single AI foundation model trained on data from multiple customers and use cases to improve picking performance and generalization across different robot hardware.

“We have set out to build the robotics foundation model, like essentially, think of it as a single brain that can be fitted on different kinds of robot hardware that can be picking, handling different kinds of items. And it's a single brain that power all these different use cases, these different hardwares, these customers.”

Engineering & simulationCovariant

Covariant leverages a flywheel where production data from deployed robots is used to train larger models with stronger emergence capabilities, allowing the system to tackle more customer problems at higher performance rates.

“That way more data allow us to train larger model that have stronger emergence capabilities that then allow us to tackle more customer problems at a higher rate of performance. that they're expecting.”

Engineering & simulationCovariant

Covariant uses simulation as a critical pillar in its development stack to generate combinatorial data and test conditions that are difficult to replicate in the real world, complementing real-world data collection.

“there's so much value you can get from simulations, um, in terms of ease of testing, in terms of ability to generate combinatorial amount of data that satisfy like the kind of conditions that you want, you have to use simulation…”