Listen to this article

Using robots to unload freight on receiving docks isn’t a novel idea. A quick YouTube search will return a slew of results. But many of these previous approaches have been stifled by high costs and the inability to handle a growing variety of packages.

Of course, robotics companies are still working on the problem, which is one of the major issues for warehouse operators. Dextrous Robotics, a Memphis-based robotics startup, is developing its Chopstick system it hopes can unload freight with more success and at a fraction of the cost of previous solutions. Co-founder Evan Drumwright discussed the approach in last week’s RoboBusiness Direct session “Advances in Robotic Picking, Grasping and Manipulation.”

Now Boston Dynamics is throwing its innovative hat into the ring. It revealed its newest robot, Stretch, a mobile manipulator designed to move boxes out of trucks and around warehouses. When it goes on sale in 2022, for a yet to be named price, Stretch will initially focus on truck unloading and later add palletizing to its repertoire. Stretch is currently being tested by a few partners. Boston Dynamics is seeking customers to pilot Stretch with truck unloading tasks.

Next-gen Handle

Stretch is the next generation of Handle, a robot Boston Dynamics introduced in 2017 that combined wheels and legs. Stretch doesn’t have legs, but it does have an omni-direction mobile base with four independently controlled wheels, a custom 7-DoF industrial robot arm that can lift up to 50 pounds, a custom suction gripper and much more. The air system for the gripper is onboard the mobile base. Stretch comes with an 8-hour battery life, but there will be a 16-hour battery option and the ability to plug Stretch in for continuous power.

Stretch uses the Pick vision system, which Boston Dynamics acquired when it bought Kinema Systems in April 2019. Pick uses high-resolution 2D and 3D vision and machine learning algorithms for robotic depalletizing. Kevin Blankespoor, Boston Dynamics’ VP of product engineering, said Stretch can pick up to 800 cases per hour. Its arm has a seven-foot reach and can reach boxes up to 10 feet high. Stretch weighs 2,650 lbs.

Stretch is semi-autonomous, Blankespoor said, and the level of autonomy depends on the application. When it comes to unloading a truck, for example, a person still needs to open the truck door, verify the content and move the Stretch robot into position by driving it inside the truck with a joystick. “At that point you hit ‘Go’ and Stretch will fully autonomously do the rest of the job unloading the boxes,” Blankespoor said.

Building pallets will be the second application for Stretch, and Blankespoor envisions more autonomy in that task. “Stretch will be navigating the aisles of the warehouse. It’ll be going to different pallets that are single-SKU pallets and grabbing a few boxes building up an outgoing pallet. For that task, Stretch will take on more of the kind of localization and navigation you see in some other AMRs [autonomous mobile robots] that navigate through the warehouse and have more autonomy.”

Blankespoor said collaborative robot arms lack the speed and strength to lift the number of heavy cases per hour that Boston Dynamics targeted. So it built a custom industrial robot arm. For safety, Stretch uses speed and separation monitoring. “As people get closer to the robot, it will slow down,” he said. “If they get closer still, it will stop. And that way we can allow the arm to move heavy boxes fast without risk of hurting anyone.”

Hello Robot also has a mobile manipulator named Stretch that is designed as a research platform. According to IEEE Spectrum, Hello Robot co-founder Aaron Edsinger, who also worked on robotics at Google while Google owned Boston Dynamics, isn’t thrilled. “We’re disappointed they chose our name for their robot. We’re seriously concerned about it and considering our options.”

Handling variety the key

As you might notice in Boston Dynamics’ videos of Stretch, all the boxes inside the truck and on the pallet look identical and are perfectly stacked. Most, if not all, trucks that need to be unloaded in the real world won’t look like that. So Stretch’s success, like other attempted solutions before it, could be determined by its ability to handle a variety of packages.

“That’s an astute observation,” said Blankespoor. “It’s pretty easy to detect a box sitting by itself. But detecting a box when they’re packed tightly together or when there’s a big variety, is a much harder task. But that’s where our Pick box detection software comes in. And that’s one of the things you’ll see next with Stretch. We’re starting off with more straightforward trucks, but we’re going to get into more complex types of boxes, different types of boxes over the coming year.”

FedEx’s Aaron Prather discussed during his recent RoboBusiness Direct keynote that the biggest challenge in solving the truck offload issue is the high level of package diversity a solution would need to be able to handle. Prather started his FedEx career offloading trucks while still in college.

“When unloading a bulk loaded truck, you may start with boxes of a certain size, then you hit a layer of bigger or smaller boxes, then a layer of rugs (because people buy rugs online now), and then back to the same size boxes you started with. This is why it is so hard to automate this process.”

Stretch robot just the beginning?

Boston Dynamics in 2020 acquired a boatload of warehouse automation patents from X Development, LLC, an R&D organization founded by Google. Boston Dynamics was owned by Google from 2013-2017. Certainly some of the IP helped with Stretch, but it likely points to a broader interest in the logistics automation space.

Industrial Perception, a company acquired by Google in 2013, partnered with Wynright on truck unloading research. Here’s a video of the partnership in action.

“Well, we’ve definitely had interest in the warehouse space since Atlas. That’s when we figured out what a big market this is, and how much potential there is for mobile robots in the warehouse space. I don’t want to say too much about the patents, but we have a warehouse robotics team that is growing rapidly, which I’m heading up. And that includes development of the Stretch robot, we have our Pick vision system, we have a fleet management software system that can pull together different robots into doing coordinated activities. So it really just fits into the kind of broader picture that Boston Dynamics is going hard into the warehouse space.”

Learning from Spot

Boston Dynamics introduced in February 2021 Spot Arm, a robotic arm for the Spot quadruped that offers a max lift capacity of 24.3 lbs. While Spot Arm looks vastly different from Stretch’s industrial robot arm, Blankespoor said the core technological building blocks are the same.

“If you look at the wrist joint of Stretch, it’s actually the same as the hip joint on Spot. So it uses the same electric motors and the same gearboxes, same sensing even the same software,” he said. “We use the same cameras and depth sensors across our different platforms. More importantly, we use the same software underneath to actually understand what those sensors are seeing – everything from box detection to obstacle detection. So even though it looks really different from Atlas or Spot, Stretch is built from most of the same technologies. And that enabled us to get a prototype built really fast.”

Atlas remains an R&D platform, while Boston Dynamics has sold more than 400 Spot robots since the quadruped first went on sale last year. Boston Dynamics won a 2020 RBR50 Innovation Award for Spot as it jump-started the commercial quadruped market.

Fleet management, WMS integration

Boston Dynamics and OTTO Motors released a video in March 2020 that demoed their vision of the future of logistics automation. In the video, Boston Dynamics’ Handle robot picked boxes and built pallets on top of the OTTO 1500 heavy-duty AMR. Blankespoor said the two companies continue to collaborate, and it’s played an important role in the development of a fleet manager for Stretch.

“Part of the important piece of that demo with OTTO Motors was to get heterogeneous robots working together. There’s a lot of AMR companies that control their own robots, and they might have a fleet manager to do that. But there’s not a whole lot of fleet managers that can control a variety of different types of robots, coordinated to do different tasks. One of the reasons we’re working with OTTO is to coordinate the pallet movement and the case movement with different robots.”

Blankespoor also said integration with warehouse management systems is on Boston Dynamics’ roadmap. But he said it’s more crucial later on when Boston Dynamics starts to tackle order building.

“Truck unloading is nice because it’s a little more isolated in terms of the integration tasks. Once the robot is in front of the truck and you say ‘Go,’ our box detection system will tell Stretch where all the boxes are, what size they are, we actually even weigh the boxes as we’re first lifting them up, so we can move them as quickly as possible. This means we don’t necessarily have to do as much integration work.”

“The second task that we’re going to bite off is order building. And for that you do need warehouse management system integration, but it should be akin to what you do with people, right? Here’s an order, go build me a pallet with five of this box and four of this box, and stack it all up. We’ll need to convey that information to our robots so they can go do those tasks.”

Evolution of a box-moving robot

You might think a logistics robot is out of left field for Boston Dynamics, but it’s shown interest in the space for a while. In 2016, one of its first videos of the next-gen Atlas robot showed the humanoid bend down, pick up a box, and place it onto a cart. Then in 2017, Boston Dynamics introduced Handle, a hybrid robot that combined wheels and legs. Stretch is the next generation of Handle.

“We got a lot of interest [for Atlas] from warehouses,” Blankespoor said. “We knew we could design a simpler robot, and that’s where Handle really came from. Handle was an offshoot of Atlas.

“We branched off and designed Handle for a couple of reasons. One was that we wanted to do something more purpose-built for the warehouse. And another reason was that we always wanted to combine wheels and legs. Handle was an opportunity to explore both of those things. So we built a couple versions of Handle and started doing warehouse tasks.”

“First we were doing pallet building, and that was working pretty well. Then we started doing truck unloading with Handle. And at that point, it was an eye-opening experience – Handle could do the job. It could grasp the boxes and move them, but it took too long. And to make a cost-efficient robot for customers, you need to move cases pretty fast. Stretch can move cases about five times faster than Handle, so that’s one of the reasons we made the next jump.”

Hyundai an early customer?

Hyundai Motor acquired a majority stake in Boston Dynamics from Softbank in December 2020 for about $880 million. A Softbank affiliate retained the other 20%. Spot Arm was an important announcement since it effectively turned Spot into a mobile manipulator, not just a data collection platform. But the introduction of Stretch is the biggest announcement from Boston Dynamics since the acquisition. So will Hyundai be an early test customer for Stretch?

“They absolutely could be,” said Blankespoor. “We’re excited to work with Hyundai. Obviously they’re a world-class manufacturer. Hyundai has its own logistics companies in Korea, and they definitely could be a customer for us in the future.”

Editor’s Note: We will have more about Stretch from our conversation with Kevin Blankespoor later this week on The Robot Report Podcast.