Explore our R&D with the world’s most dynamic humanoid robotRead More
Discover the past innovations that informed our current productsRead More
Meet the team behind the innovationsRead More
Learn how we develop and deploy robots to tackle industry’s toughest challengesRead More
Start your journey at Boston DynamicsRead More
Stay up to date with what we’ve been working onRead More
Discover the principles that guide our work and policiesRead More
Warehouse Robotics
Blogs •
Emerging from years of experimentation and R&D, the Stretch robot offers the mobility, power, and intelligence to become the future of warehouse automation.
Compared to the superhuman mobility of Boston Dynamics’ humanoid robot, the robot currently disrupting the warehouse industry can look positively boring.
Stretch is one of the first commercially available robots designed to tackle one of the most tedious and strenuous jobs in the warehouse—unloading boxes. While less flashy than Atlas, Stretch is no less superhuman, built on a shared technical foundation and offering powerful functionality and cost-effectiveness for warehouse operations.
Stretch helps employees avoid injuries prevalent in the work of manual unloading. They can then shift their attention to other tasks within the warehouse, such as quality control and order fulfillment, that benefit from human reasoning and dexterity.
“For a company that has made its history with exciting robots walking around and doing back flips, you get a very different sense when you see Stretch operating,” said Kevin Blankenspoor, SVP of Research Collaborations at Boston Dynamics, who formerly led warehouse robotics. “It’s really interesting for a few minutes and then you realize it’s boring. This thing just keeps going all day long and doing real work.”
Stretch rolls around warehouses on a mobile base with a footprint about the size of a pallet. It has a flexible arm with a vacuum gripper that enables it to deftly remove boxes of many shapes and sizes from shipping containers and tractor-trailers.
Stretch uses multiple sensors to recognize objects in its surroundings. Machine learning algorithms process all this data and help Stretch decide what to do next. Making decisions in real time is important in warehouse work, where environments and boxes vary often. Stretch is poised to take on the monotonous, back-breaking work of unloading containers for multiple shifts, seven days a week. These advanced behaviors and benefits have led companies like DHL, Maersk, and H&M to integrate Stretch into their daily inbound.
“Stretch is out in the world moving boxes every day at multiple sites,” said Mike Murphy, Stretch Chief Systems Engineer. “We hear stories of trailers pulling up full of thousands and thousands of 50 pound boxes. This could be in the summertime where the temperatures are over 100° inside this metal box, and those are grueling conditions for the people who are asked to unload them. They’re very happy to be able to push a button and have Stretch go in and do that job while they work on other higher value operations within the warehouse.”
While the Stretch robot of today doesn’t bear a strong resemblance to the recently announced commercial version of Atlas, they both grew out of the same initial research into mobile manipulation. In fact, the idea of engineering a warehouse robot only came after customers responded to a 2016 video of an Atlas robot lifting and carrying boxes. A new market came calling.
“We were doing some internal research on manipulation with Atlas, and we happened to use boxes for this demonstration,” said Murphy. “When we released this video, we got a lot of incoming calls that Atlas would be perfect for a warehouse and, ‘We want to buy one.’ We knew that it wasn’t really the right fit for a product at the time but we realized that was an indication of a market need.”
The road to Stretch involved plenty of trial and error. Engineers named their first concept robot Handle. They used the torso and arms of Atlas but gave Handle wheels instead of feet for navigating warehouse floors efficiently, and legs to extend its vertical reach. In true Boston Dynamics style, the team put Handle through rigorous testing during a three-year research phase, from 2016 to 2019.
“We wanted to understand the capabilities and limits of a system that has that combination of wheels and legs,” said Murphy. “Early versions of Handle leveraged copies of Atlas’ arms on the torso to pick up and move objects, but they weren’t really designed for the kinds of range of motion and applications that Handle was really intended to do.”
For Handle 2.0, the design team switched from the two arms to a single, high-strength arm with a vacuum suction gripper on the end. They also moved the top-heavy torso—which contained the battery—to a new tail section that acted as a counterbalance, giving Handle the appearance of an ostrich on wheels. During tests, Handle could careen down staircases and do figure 8s, but it still wasn’t optimal for picking up and moving boxes. The robot required too many movements to balance and get into position, and could only see where the gripper was pointed.
“Imagine trying to do certain things in a warehouse environment when you’re wearing roller blades,” said Arash Ushani, Machine Learning Lead for Stretch. “It just slows down the speed of operation when you’re trying to move a box from point A to point B. It’s hard to continually have to turn.”
Conceding that it needed a brand new design optimized for moving boxes in warehouses, the Handle team pivoted in 2019 to its final design, called Stretch.
“We reimagined the entire design into something that really suits the workspace of the applications in the warehouse,” said Murphy. “We revisited every component to make sure it was going to be reliable, cost-effective, safe and have the performance needed for the market. What resulted is a brand new machine.
Looking more like an industrial robot, Stretch doesn’t resemble a human or animal. It has a sturdy wheeled base that moves smoothly across industrial floors in any direction.
“Maneuverability was really a big advantage, especially in tight spaces, like the back of a truck or narrow aisles in a warehouse,” said Ushani.
Atop the base, Stretch engineers installed a 7 degrees of freedom robotic arm. Among traditional 6-axis arms, Stretch’s arm stands out by enabling grasping from various angles and orientations, as well as smoother and faster case handling motions. This allows Stretch to work in tight spaces without the arm links hitting objects like racking or container walls. The team also updated the vacuum gripper with sensors and pneumatic controls that enable Stretch to grip boxes weighing up to 50 pounds, including those with some damage. For good measure, Stretch also has a new, quieter onboard vacuum pump.
“There’s a little more flexibility than a traditional industrial robot arm in order to fit into tighter spaces and have more solutions to move through the space,” said Murphy. “At the end, we have a vacuum gripper that allows us to attach to the box and maintain a good grasp.”
Beyond the mechanical improvements, what sets Stretch apart is its ability to use machine learning to adapt to its surroundings. Stretch doesn’t blindly operate—it actively interprets its environments and makes decisions in real time about which case it will pick next and how it will do so.
“Stretch is continually learning and improving,” said Ushani. “We’re constantly analyzing data that we’re getting in for moments in the field. Machine learning, or especially deep learning, is inherently very data hungry, so the more data you can collect, the more images you can capture, the more experience Stretch can leverage, the better off it’s going to do.”
A swiveling “perception mast” equipped with cameras and sensors gives Stretch a 360° view of its surroundings. The mast is individually actuated and moves independently from the arm. As a result, even while Stretch is busy lifting and placing a box on a conveyor belt, it’s already using its cameras to determine the next box to grab.
“With the sensors on the perception mast, we can point at anything we need to see—whether it’s boxes, pallets, or conveyors—and we can opportunistically take pictures whenever we need to,” said Ushani. “Those pictures are then fed into a custom-built, machine learning algorithm that’s able to identify where the objects of interest, like boxes, are in the image.”
This evolution of Stretch was guided by close collaboration from a small group of early adopter customers. Their feedback helped ensure that we were developing the right solution for the right challenges. What started as a response to an unexpected market need has evolved into a thriving solution on the cusp of large-scale adoption. Stretch has already moved millions of boxes since its 2023 release, and the robot is gearing up to go global.
This R&D journey doesn’t stop with a single product. We are continuing to develop new capabilities for Stretch and expand the total ecosystem of software, service, and support to enable customer success. And now we’re repeating the process with our new electric Atlas to solve manipulation challenges in other industries. We’re committed to finding practical solutions to the toughest real world challenges.
“It’s really cool to know that the things we’re creating are out in the world helping people,” said Murphy. “And as we scale up, there’s probably a Stretch running at all times of the day.”
Recent Blogs
•8 min watch
The Evolution of Stretch
•5 min read
What It Takes to Pick a Million Boxes
•4 min read
Taking Mobile Case Handling Beyond the Container