24.9 C
New York

Amazon’s Vulcan robots now store items faster than humans

Published:

Today, at an event in DortmundAmazon announced the launch of a new robotic system named Vulcan. The company describes it as “its first robotic device with a true sense of touch” and says that the system is designed to help (or take over depending on your perspective) in stowing and selecting items in its mobile robot inventory system.

Two upcoming papers will be published in
IEEE Transactions on Robotics (19659004), Amazon researchers describe both the stowing side and the picking side of the system. In a previous interview, we discussed stowing in depth.

Aaron Parness
is the director of applied sciences at Amazon Robotics. Parness and his team made significant progress in stowing, improving speed and accuracy over more than 500,000 stows. The average stowing robotic is now slightly quicker than the average human. Parness was contacted by Wto discuss the latest developments in stowing and how Vulcan manages picking. This separate article contains more information. It’s an entirely different problem and worth reading.

Optimising Amazon’s Stowing

Stowing refers to the process Amazon uses to bring products into its warehouses, and add them to its inventory for you. Amazon has optimized this process to maximize space and time efficiency. Human towers are presented with an

Mobile robotic pod
with fabric cubbies and elastic bands on the front to prevent things from falling out. The human’s task is to find an empty space in the bin, remove the plastic band, and place the item there. The new home of the item is recorded in Amazon’s system. The pod then drives back to the warehouse and the next pod arrives, ready for the following item.


Different manipulation tools are used in order to interact with human optimized bins.

Amazon

This new paper on stowing contains some interesting statistics about Amazon’s inventory handling process that help put the problem into perspective. Amazon warehouses stow more than 14 billion items by hand each year. Amazon hopes that Vulcan robots can stow 80 per cent of these items, at a rate 300 items per hour while operating 20 hours a day. It’s an extremely high bar.

Amazon’s robots have become quite adept at stowing after a lot practice. Parness says that the stow systems are now operating three times faster than they were 18 months ago. This means that it is actually a bit faster than an average person. Parness says that robots are still beaten by experts. “The fastest humans are like Olympic athletes. They’re faster than robots and can store items in pods with much higher density. This is especially important in urban areas, where space is limited. Humans are very creative in this three-dimensional physical ”

Tetris
is still being worked on by the robots.


Robots excel at planning ahead. This is why the average robot tower is now able outpace the human tower–Tetris is also a mental process. Parness says that robots can use a lot more data than humans to optimize where and when things are stored. “When you are a human doing this task, your buffer is 20-30 items. You’re trying to find a way to fit them into different bins and remembering which item goes where. The robot can see all the properties of each item at once and also the bins of the next two pods. We can optimize the entire set of information within 100 milliseconds.”

In essence, robots are better at optimizing the planning side of Tetrising while humans are still better at manipulating the pieces. However, this gap is closing, as robots become more adept at working in clutter and with contact. Amazon has been using Vulcan stowing robotics for over a month in real-life warehouses in Germany, Washington State and other countries to collect data. These robots have successfully stowed thousands of items.

Stowing, of course, is only half of what Vulcan was designed to do. Parness’s in-depth discussion on picking also presents unique challenges. You can read it here.








www.roboticsobserver.com

Related articles

spot_img

Recent articles

spot_img