Logistics is in the midst of a technological revolution. The best estimate is that fully autonomous semis will be on US interstates at scale by 2028. It is not hard to see a future in which very few human drivers will be on the road.

But what about the warehouse? How close are we to a dark warehouse? A dark warehouse is one that employs no one to put goods away, pick them, or stage them at the dock. A dark warehouse would just have a few engineers to keep the machines running.

A warehouse that stores and moves pallets using an automatic storage and retrieval system is already fully automated. Meijer, the midwestern grocery, has such a warehouse . Each picking, picking just one or a few items to fulfill an order (often an ecommerce order), is the bigger challenge when it comes to creating a dark warehouse. Ecommerce warehouses are generally far bigger and employ far more workers than warehouses focused on case or pallet fulfillment.

To answer this question, I talked to John Santagate. John was the senior vice president of global robotics at Infios, formerly known as Körber Supply Chain Software. He started a very successful warehouse robotics practice at the company. He left Infios about a year ago and started his own consulting business in this domain. He is on the advisory board of Radiant Robotics and Instock.com .

While this is technologically possible, Mr. Santagate pointed to the prevalence of cobot solutions, which will limit the scale at which it will be done.

North American companies ordered nearly 18,000 robots valued at about $1.2 billion in the first half of this year, according to a recent report from the Association for Advancing Automation. In terms of units, a large percentage of those robots are cobots – robots that work in concert with human pickers.

One of the best-known, and probably the largest cobot provider, is Locus Robotics . With Locus, bots move to a pick location, and a worker goes to the bot and picks onto it. Then, the worker may be asked to move ten feet down an aisle where another robot is waiting. The worker goes to that bot and picks onto it. This optimization logic is designed to reduce pickers’ travel time while ensuring orders are packed and ready to ship on schedule.

These solutions, sold on a Robotics-as-a-Service model, deliver strong return on investment. Picking rates increase by 2-3 times. Furthermore, the cobot solutions scale in a way that bolted-down automation does not. If demand increases, it is not difficult to add a few more bots. Scaling up bolted-down automation is a much more difficult endeavor.

Another type of robotic solution is tote-to-worker. The mobile warehouse robot revolution was kicked off when Amazon bought Kiva Systems in 2012. These small, low-to-the-ground robots, used for e-commerce picking, get under a small, square rack with multiple slots. The robot lifts the rack, brings it to the worker, then spins it so the correct side faces the picker. Then the slot to be picked lights up and a person makes the pick. This is goods-to-man automation.

There are other versions of this that accomplish the same things in different ways. An AutoStore system comprises a structural grid, storage bins, robots, and ports (pick-and-put-away stations). Products are stored in stacked bins within grid. As a stock keeping unit is needed to fulfill an order, the control system alerts a robot to find the bin containing the item. If the bin is buried beneath other bins, a robot places the extra bins on top of the grid until it retrieves the desired bin. It then takes that bin to the edge of the grid and delivers it to the operator at a workstation, who completes the pick. As the bin is delivered, another robot on top of the grid returns the other bins to their original storage location.

But with these goods-to-man solutions, you still have people at workstations completing the pick. Can picking be fully automated?

A few years ago, mobile robots came to market that had picking arms on top. This was slow and cumbersome. Ultimately, this was a narrow use case that made sense only for high-value goods or controlled substances. Humanoid robots have gotten a lot of press recently. But “they’re too slow,” Santage exclaimed, “they’re too expensive.” With humanoid picking/traveling bots, there are use cases, but they remain limited.

Nevertheless, great progress is being made. Santagate pointed me toward a post Gina Chung, the chief strategy officer at Locus, had made on LinkedIn. “Robotic hands are mesmerizing,” she said. “But the challenge is that warehouse fulfillment doesn’t just require dexterity; it requires industrial speed. Piece picking demands both.”

Locus Robotics acquired Nexera Robotics in May. This solution uses suction grasping. In the video, picking is done very quickly, quicker than what a human could do. The recommended weight range is items weighing under 5 pounds.

So, to sum it up, Santagate said, “I’m confident we’re at a place where we can deliver on the touchless pick. Induct (goods) on one side to a tote command system, whether that’s an Autostore” or some similar system. “The technology all exists to do that.”

But just because something is possible does not mean it is wise or that it makes financial sense. If you’re designing a greenfield system and not retrofitting an existing system, “that’s where I think that there’s some growth opportunity” for the fully automated warehouse.