Amazon touts warehouse robots but humans still better at tasks

In a couple of research papers, retail giant Amazon says that its warehouse robots can now pick and stow products. But people are still better and can’t be replaced, the tests have shown.

With humans, the way things work in an Amazon warehouse is pretty straightforward. Workers check the quality of newly-arrived items and stow them in storage pods.

After an order is placed, they need to move the entire storage pod to a picking station where other workers – again, humans – remove the purchased item and send it for packaging and shipment.

It’s quite simple indeed, so Amazon thought that automating these tasks wouldn’t be too hard. The company has now revealed it’s been testing robots called “Stow” and “Pick.”

The “Stow” features a pinch gripper, an extendable plank for manipulating items in storage bins, a visual perception system for assessing available space, and a machine learning model to predict the packing success rate.

The “Pick” robotic system, designed for autonomous picking of targeted objects from “cluttered and deformable shelves,” is supposed to deal with a critically important task in Amazon warehouse operations.

The company has detailed early results of testing both systems in research papers, and they seem promising at first glance.

For instance, the “Stow” attempted to move over 500,000 items during a test and did its job properly about 85% of the time.

“The system achieves human levels of packing density and speed while prioritizing work on overhead shelves to enhance the safety of humans working alongside the robots,” Amazon says in the research paper (PDF).

But here’s the catch. The paper also says: “Over the month of March 2025, humans stowed at an average rate of 243 units per hour (UPH) while the robotic systems stowed at 224 UPH.”

The “Pick’s” success rate, described in another paper (PDF), is even higher, reaching 91% across 12,000 pick attempts during its six-month trial deployment in 2024 and 2025.

However, the robot also apparently rejected 19.4% of pick requests when its machine vision couldn’t recognize the item or when it refused to perform the task out of concern for item damage.

It would seem human workers are still needed. But Amazon is undeterred and says: “By sharing our experiences and insights from a large-scale, real-world deployment, we aim to encourage collaborative efforts that push the boundaries of robotic manipulation in industrial settings.”