Warehouse Automation Versus General Manipulation

Diving deeper into

Mimic Robotics

Company Report
Traditional warehouse automation companies like Berkshire Grey and Plus One Robotics target similar manipulation challenges but with more specialized, task-specific approaches rather than general-purpose foundation models.
Analyzed 10 sources

The key divide is not whether these companies can pick things up, it is where the intelligence sits. Berkshire Grey and Plus One sell warehouse workcells tuned for narrow jobs like parcel induction, depalletizing, sortation, and piece picking, while Mimic is trying to build a reusable manipulation model and hand that can transfer across many tasks. That means the incumbents optimize for immediate throughput and ROI in one workflow, while Mimic is optimizing for broader capability and learning.

  • Berkshire Grey packages robotic picking and sortation as modular systems that drop into existing fulfillment lines. Its materials emphasize concrete operating metrics, like 2,000 plus units per hour per module, 30% higher volume, and 70% lower labor in retailer deployments. That is classic warehouse automation, solve a known bottleneck with a purpose built cell.
  • Plus One is similarly specific. Its products focus on induction, palletizing, and depalletizing, with PickOne vision software and Yonder remote supervision handling edge cases. In practice, that means a robot arm grabs parcels off a chute or pallet, and a human steps in remotely only when the system gets stuck. The company says it has executed over 1.5 billion picks across 15 countries.
  • Dexterity sits between these worlds. It is more AI heavy than classic warehouse integrators, but it still sells turnkey systems for high volume warehouse tasks rather than a broadly transferable foundation layer. The broader humanoid and robotics foundation model wave matters because it is pushing down component costs and reframing dexterity as a general software problem, not just a warehouse cell design problem.

The next phase is a collision between task specific incumbents and general manipulation platforms. If foundation models become reliable enough on real warehouse floors, value will shift from selling one more induction or sortation cell to owning the learning loop that can unlock many workflows on the same hardware stack, and warehouse automation starts to look more like a software scaling business.