BotsGenius
ENNL
← Back to all companies
United StatesAI & softwareCommercial

Covariant

Foundation models for robotic picking and manipulation.

Covariant is an Emeryville-based company building AI foundation models for robotic picking and bin-picking in warehouse and logistics environments. A significant portion of its team was subsequently hired by Amazon.

Covariant's core thesis is that a single, large foundation model trained on diverse manipulation data can outperform task-specific robot programs in real warehouse deployments. Its RFM-1 (Robot Foundation Model) is trained on a broad range of pick-and-place demonstrations and generalises to novel objects without per-SKU programming. The company deployed its system at logistics operators across North America and Europe before much of its research team transitioned to Amazon, marking a significant consolidation of manipulation-AI talent within the e-commerce giant. Covariant's work has been influential in demonstrating that transformer-based policies can work reliably in high-throughput industrial settings.