Amazon pushing envelope on warehouse automation

Listen to this article

Amazon’s newest robot, Sparrow, can pick individual items for packaging. | Source: Amazon

Amazon continues to be a leader in automating warehouse operations, and that includes not only using existing technology but developing its own products and buying manufacturers of interest. In September, the company announced it was acquiring Cloostermans, a Belgium-based company that specializes in mechatronics.

Cloostermans has been selling products to Amazon since at least 2019, including technology Amazon uses in its operation to move and stack heavy pallets and totes and robots to package products for customer orders. Amazon said this acquisition will ramp up its R&D and deployment in those areas.

“We’re thrilled to be joining the Amazon family and extending the impact we can have at a global scale,” said Frederik Berckmoes-Joos, CEO of Cloostermans. “Amazon has raised the bar for how supply chain technologies can benefit employees and customers, and we’re looking forward to be part of the next chapter of this innovation.”

The financial terms of the deal were not disclosed. Founded in 1884, Cloostermans has been privately held for the last six generations. Cloostermans has about 200 employees that will join Amazon Global Robotics’ growing presence in Europe.

Amazon said it has deployed more than 520,000 robots in its facilities worldwide. Of course, that all started in 2012 when Amazon acquired Kiva Systems and its automated guided vehicle (AGV) technology for $775 million. Amazon recently introduced its first autonomous mobile robot (AMR), Proteus. It does similar tasks as the Kiva robots but can work freely around Amazon workers instead of caged areas.

Since the acquisition of Kiva Systems, Amazon has built out an impressive robotics portfolio. And it’s not limited to just warehouse robotics systems. Amazon has a pending deal to acquire iRobot for $1.7 billion that is being reviewed by the Federal Trade Commission. It also acquired Canvas Technology, Dispatch and Zoox, and invested in companies such as Agility Robotics, which demoed its bipedal robot, Digit, at RoboBusiness 2022.

“Amazon’s investments in robotics and technology are supporting how we build a better and safer workplace for our employees and deliver for our customers,” said Ian Simpson, vice president of global robotics at Amazon. “As we continue to broaden and accelerate the robotics and technology we design, engineer and deploy across our operations, we look forward to welcoming Cloostermans to Amazon and are excited to see what we can build together.”

Teaching robots to do more

Robots picking items in Amazon’s warehouses need to be able to handle millions of different items of various shapes, sizes, and weights. Right now, the company primarily uses suction grippers, which use air and a tight seal to lift items, but Amazon’s robotics team is developing a more flexible gripper to reliably pick up items suction grippers struggle to pick.

Amazon is teaching robots how to understand cluttered environments in three dimensions, locate specific items and pick them using a pinch grasp, or a thumb and finger hold. The company’s current vacuum-like grippers use elastic suction cups that form to the surface of an item. This creates a tight seal that allows the robot to pick objects.

Amazon said this method works great for flat items that only require one point of contact for picking, like rulers or cards. It’s less effective, Amazon said, for items that require more than one point of contact to pick up, for example, a book will fly open if you pick it up from just the front or back cover.

Suction grippers also struggle to get a tight seal on bags filled with granular items, like marbles, according to Amazon. And even on items these grippers can pick up well, if the angle of attachment changes because of the momentum of the robot arm swinging it from one place to another, then the seal will break too early and the robot drops the item.

These cases are why Amazon is interested in the pinch-grasp method. Despite how natural it is for humans, it’s not a simple one to develop in a robot. To teach a robot to pick items out of piles of other items using this method, researchers first needed to teach it to be able to estimate the shape of items that could be partially obscured by other items.

As humans, we do this without even thinking about it, but robots have a much harder time understanding the whole shape of an item if they can’t see all of it. Amazon’s robots gauge what they’re picking by using multiple camera angles and machine learning models trained to recognize and estimate the shape of individual items. The robot uses this to decide how to best grasp the item on two surfaces.

Once the robot makes those observations, it uses a set of motion algorithms to combine the information it gathered about the scene and the item with the known dynamics of the robot to calculate how to move the item from one place to another.

The robot also continues to use its multiple-angle view of the situation throughout the pick. This is another deviation from typical picking methods, where a robot won’t usually continue to look at the scene as it carries out a pick. So far, Amazon’s team has seen encouraging success with its pinch-grasping robots. A prototype robot achieved a 10-fold reduction in damage on certain items, like books, without slowing down operations, Amazon said.

Despite this, Amazon still sees room for improvement. The team is currently using an off-the-shelf gripper that can only pick items that weigh less than 2 lbs. This makes the gripper capable of handling only half of the items that Amazon has available for purchase. Going forward, the team plans to design its own gripper for the job.

In the future, Amazon hopes that it can implement its pinch-grasping robot alongside its current suction ones so that it can decide which robot would be best suited to picking each individual item. The company is using a similar strategy with Proteus.

Creating an ecosystem

Earlier this year, Amazon unveiled Proteus, its first-ever AMR. The company first entered the mobile robot space in 2012 when it acquired Kiva Systems, which offered automated guided vehicles (AGVs) that have been at work in Amazon’s warehouses since.

Proteus has a similar design to the Kiva robots. It slides under Amazon’s GoCarts, lifts them up and moves them across warehouses to employees or other robotic cells. Unlike the Kiva robots, which currently operate in caged-off spaces away from Amazon employees, Proteus is able to work freely among them.

This change means Proteus needs to be prepared to adapt quickly to unexpected changes in its environment. John Enright, principal engineer at Amazon Robotics, recently gave some insight into how the company developed the technology behind Proteus. He explained the approach to navigation in the video above.

“Our design focuses on safety, efficiency and cost-effectiveness,” Enright said. “We employ a wide range of diverse and redundant sensing modalities that allow us to provide certain guarantees on vehicle behavior.”

Proteus’ job is to store, move and sort Amazon’s blue GoCarts, a central part of the company’s logistics operations. The AMR travels to where the carts are and slides underneath them to move them. It uses general navigation abilities to travel to the general location of the GoCarts, and then uses its high-precision LiDAR to find the carts.

To slide under the cart, Proteus uses a two-step detection and motion process. First, the robot will perform a small “S” curve to remove any lateral error in its positioning under the GoCart. Next, it performs a straight motion to tunnel under the cart and lifts it.

Proteus carries the cart to its desired storage location, which it identifies with Amazon’s fiducial plus. Fiducial plus is a custom-made ground target that aids Proteus in its alignment capabilities and finding storage cells. These fiducials help the robot to perform millimeterlevel corrections on its positioning. The AMR has been deployed in Amazon’s outbound GoCart handling areas in its fulfillment and sorting centers. A source told The Robot Report Amazon will use both the Proteus AMRs and the Kiva-like AGVs moving forward.

Amazon also recently unveiled Sparrow, a robotic arm capable of picking individual products before they get packaged. Unlike Amazon’s Robin and Cardinal robots, which pick and organize packages to be sent out for delivery, Sparrow can handle individual products. This isn’t a simple task in a place like an Amazon warehouse, where over 100 million different items could need to be processed. Sparrow can pick 65% of them, according to the company. Sparrow can pick a variety of items, like DVDs, socks and stuffed animals, but struggled with items that have loose or complex packaging.