Packing online orders into boxes for shipping—so-called picking—has been a major area of job growth in recent years. Warehouse fulfillment now employs nearly a million Americans, helping offset massive job losses in brick-and-mortar retail.
But those warehouse workers may soon be under threat themselves, as engineers get closer to developing robots that can do the job more cheaply and reliably.
The complexity of sorting through hundreds of thousands of different products has made automation impractical so far. But according to a new report in the Wall Street Journal, machine learning is now making the robots more flexible and effective.
Get Data Sheet, Fortune’s technology newsletter.
Some of them are already inching into service. A subsidiary of robotics company Kuka sold its first picking robot in the U.S. to an unnamed retailer early this year, and the startup RightHand Robotics is testing its picking robots at a distribution center of the retailer Hudson’s Bay.
But experts speaking to the Journal say full commercialization of automated picking is at least a year away. Amazon, the 800-pound gorilla of e-commerce, is still funding research and competitions to develop the technology further.
Robots are already being widely used by Amazon and other retailers in warehouses to bring shelves holding products to human pickers so they don’t have to waste time walking around what are often huge facilities. However, those robots do not actually pick and pack products for shipping.
When picking robots do arrive in full force, picking robots could cut the labor cost of online order fulfillment by one-fifth, according to a consultant who spoke to the Journal. That would make companies like Amazon more profitable. But it could also fulfill some of the worst anxieties about automation and labor by pushing huge numbers of workers back out into the cold.
