Skip to main content

Amazon’s latest robot champion uses deep learning to stock shelves

Amazon’s latest robot champion uses deep learning to stock shelves


But it's not working in warehouses just yet

Share this story

Amazon has crowned the latest champion in its robotic picking challenge — an annual competition that looks for robots that could one day work in the company's warehouses. It's basically American Idol, but for robotic arms that can grab items off a shelf and put them back again. Competitors are asked to handle a range of products, from toiletries to clothes, and then scored on speed and accuracy in stocking shelves.

The winning robot picked 100 items an hour, compared to a human picking 400

This year's contest was won by a joint team from the TU Delft Robotics Institute in the Netherlands and the company Delft Robotics (both named after the city of Delft). The team's robot managed to pick items from a mock Amazon warehouse shelf at a speed of around 100 an hour, reports TechRepublic, with a failure rate of 16.7 percent. That's slow compared to what a human can manage (around 400 items an hour), but a big improvement on last year's winner, which managed just 30 items in the same time. The team also won the stowing challenge, placing items back on the shelves.

The biggest trend in 2016's competition is the rate of improvement compared to last year's contest (the first that Amazon organized). The picking part of the challenge was much harder than in 2015, with a greater number of objects to choose from packed into more densely-filled bins. Despite this, just four teams failed to score any points at all, compared to 2015, when half of the robots failed completely. Six of the teams in 2016 also managed to score more than 40 points, which would have been enough to earn them third place in 2015.

Although the picking challenge is essentially a try-out for Amazon's own warehouses, the company insists that its intention is not to make its human workers obsolete. "Robotics enhance the job for employees but [do] not replace them," a spokesperson for the company told TechRepublic. "In fact, we continue to hire. Many of those roles are being created in buildings where employees are working alongside Amazon robotic drive units." The spokesperson added that in the company's robot-filled warehouses the end result was "a symphony of humans and technology."

amazon bought kiva robotics in 2012 for $775 million

Amazon points out that since it bought robotics firm Kiva in 2012 for $775 million, the number of employees in its warehouses (which it dubs fulfillment centers) has continued to grow. In the fourth quarter of 2014, there were some 154,100 full-time and part-time employees in the company's warehouses, compared to 230,800 employees in the fourth quarter of 2015. However, to credit this growth in employment solely to efficient robot helpers seems overly optimistic.

Amazon currently has about 30,000 Kiva robots working in its warehouses. These bots are squat, orange-colored machines which scoot underneath shelves of items, lifting them up off the floor. They're sophisticated in terms of overall strategy and movement, but their dexterity is non-existent — they're essentially clever mobile coffee tables. Picking and stowing objects on the shelves has proven tricky to automate, but Amazon's contest aims at plugging this technology gap.

In the case of the team from Delft, the answer was artificial intelligence. The researchers used deep learning techniques — which can churn through vast amounts of data to look for recurring patterns — to analyze 3D scans of the objects their robot had to pick and replace. A custom gripper-suction arm attached to an off-the-shelf Yaskawa robotic arm was used to do the actual mechanical work, while the AI-powered software gave the robot an edge over the competition.

Speaking to The Verge, Carlos Hernandez Corbato, a postdoctoral researcher who led the Delft team along with Kanter van Deurzen, said that he could imagine robot pickers showing up in warehouses in the next five years. He added, though, that "in a complex and diverse environment such as current warehouses [...] we need human operators to analyze unexpected problems," and that sorting cluttered, unorganized shelves is still a challenge. "Human intelligence and intuition is key to solve unexpected situations," said Corbato. "The most efficient work is that of a human operator with the most powerful [robot] tools."