Robots Atlas>ROBOTS ATLAS
Robotics & Hardware

Stationary OP1 humanoid handles 30% of orders at Highline's hub

Stationary OP1 humanoid handles 30% of orders at Highline's hub

US logistics provider Highline Commerce has deployed stationary OP1 humanoids from startup Ultra Robotics that now handle about 30% of daily order fulfillment at its Brooklyn hub. The robots run on Physical Intelligence's π0.6 foundation model and show that warehouse automation does not have to rely on legged, mobile machines.

Key takeaways

  • The OP1 “Operator” handles about 30% of Highline Commerce's daily order volume.
  • Design: a stationary dual-arm system with a 10-foot reach, plugged into a standard outlet, running 24/7.
  • Decision layer: Physical Intelligence's π0.6 foundation model.
  • The robots handle pick, pack, scan and conveyor tasks for over 200 retail brands.
  • Deployed at Industry City in Brooklyn's Sunset Park — two robots went live the same day.

A stationary humanoid instead of a walking one

Ultra Robotics, a Y Combinator-backed startup, went against the industry's rush toward mobile humanoids. OP1 is a pair of arms on a fixed base plugged into a standard outlet.

Its working reach is about 10 feet. Dropping legs and a battery is deliberate: it removes tipping risk and battery swaps, which allows continuous 24/7 operation with no downtime.

It is a conscious trade-off. A mobile humanoid does more in theory, but in a warehouse the most value comes from reliable, repeatable grasping and packing in one spot. OP1 bets on exactly that, rather than chasing full mobility that rarely pays off on a fulfillment floor.

A brain from Physical Intelligence

Hardware is one thing, but OP1's capabilities come from its AI layer. The robot uses the π0.6 foundation model built by Physical Intelligence — one of the leading firms building general control policies for robots. The model lets OP1 handle unpredictable packages and varied products, and recover from errors on its own without human help.

That matters because e-commerce fulfillment is a chaotic environment: different shapes, sizes and packaging. A rigidly programmed robot fails here. A foundation model trained on large manipulation datasets brings the flexibility earlier rule-based Pick-and-place: The classic robot routine: grab an object in one spot and place it in another. Traditionally it follows rigidly programmed positions, so it struggles with unusual objects. systems lacked.

Why Brooklyn, not Silicon Valley

OP1 runs at Industry City — a 35-acre tech and manufacturing campus in Brooklyn's Sunset Park. Both Ultra Robotics and Highline Commerce are based there, which lets engineers test hardware on a live production floor rather than in a lab.

The density of on-site manufacturing, fulfillment, and robotics companies creates an ecosystem that simply couldn't be replicated in Silicon Valley.

Richard Hurley — CEO, Highline Commerce.

The proximity of robot maker and customer cut deployment to hours — in one case two robots went live the same day. They handle Pick, pack, scan: Picking, packing and scanning — the core steps of fulfilling an order in an e-commerce warehouse. and conveyor tasks for over 200 retail brands operating in the hub.

Why it matters

The OP1 deployment challenges the assumption that the future of labor automation must be a mobile, bipedal humanoid. For many warehouse tasks a stationary pair of arms with a good control model is cheaper, safer and easier to deploy than an expensive walking machine. It is a pragmatic approach that could speed up real robot adoption.

The case also shows value shifting toward the AI layer. Ultra Robotics does not build its own model from scratch but buys one from Physical Intelligence, focusing on hardware and deployment. That split of roles — base-model supplier and hardware integrator — is starting to resemble the software ecosystem and may become the dominant pattern in applied robotics.

What's next

  • Scaling beyond the current 30% of Highline's volume will show whether OP1's stationary model holds reliability under heavier load.
  • Future versions of Physical Intelligence's foundation model (the π line) will directly shape how well OP1 handles unusual products.
  • The hours-long deployment model, built on Industry City proximity, will be tested on whether it can be repeated beyond a single campus.

Sources

Share this article