From DARPA to distribution centre: How Boston Dynamics went from military to warehouse operations
The firm, most famous for its walking robots intended to support military operations, is now looking to lighten the load in distribution centres
If you are aware of Boston Dynamics, it’s most likely through the viral videos of their robots that have collectively been watched hundreds of millions of times. These machines were so effective at grabbing the popular imagination because of how closely they mimicked organic movements, first as dog-like quadrupedal robots and then with the advent of their PETMAN and Atlas models, as distinctly humanoid creations.
The origins of these robots and the company comes from research and development contracts for the US military, with the Defense Advanced Research Projects Agency (DARPA) being an enthusiastic early funder, seeing the potential in their four-legged systems to help haul heavy equipment over rough terrain.
As it turns out, it wasn’t just military and defence eyes that were taking an interest in these advanced robots. Logistics operators were also watching and saw potential in the complex manipulation and movement that some these machines were capable of.
Now the company is making a transition into distribution centres, with its new robots forming the next step in an already remarkable journey.
Reuters Events, Supply Chain spoke to VP of Product Engineering Kevin Blankespoor to understand the hardware, software and ambition behind this shift to supply chain.
Taking the man out of the machine
Blankespoor says their turn towards the logistics space all started five years ago, “When we started working on our newest version of Atlas.” This, as part of their research process, conducted trials in the “fundamentals of grasping boxes… picking up heavy loads and maintaining balance and that kind of stuff.”
That's when we started visiting warehouses and understanding more about what their needs were, how big of a market it is, and some of the fundamental challenges they face
As with many of their creations, they videoed its efforts and released them online, which “Got a lot of interest from the warehouse space,” and then the journey unfolded from there.
“At that point, it was fundamentals - research and development - but that's when we started visiting warehouses and understanding more about what their needs were, how big of a market it is, and some of the fundamental challenges they face,” remembers Blankespoor. “After making the rounds and visiting lots of warehouses, we thought, oh, they're asking for Atlas [their advanced humanoid model] someday, but we think we could actually develop a robot that's a lot more simple to do the tasks that they're asking about, especially case handling and moving boxes.”
After this happy accident and exploration of the sector, they developed their prototype Handle robot to zero in on warehousing needs.
How to handle a problem?
With Atlas, they had a robot capable of running, jumping, picking up boxes and even a light session of parkour. However, this was top-of-the-line tech designed to demonstrate what could be done. Boston Dynamics engineers needed to pare things back and concentrate on the day-to-day of warehouse operations.
With this in mind, they “Designed a first version of Handle to kind of explore the robot, and then the second version really [was] designed for warehouses.” Although the Handle robot “May have the same attributes that made Atlas attractive to people in the warehouse, meaning it had a small footprint and it could pick up heavy boxes,” explains Blankespoor, “it was a lot more simple. There were just a lot fewer moving parts. It was going to be lower cost and complexity.”
Using this base, Boston Dynamics was able to explore the problems and complexities in terms of moving goods in distribution centres with real-world tests.
“The first application we did was pallet building, which looked pretty good, and the second application that we did with Handle in the warehouse was unloading trucks.” In this task “Handle could get the job done (it could get all the boxes out of the truck), but it took a while” due to Handle’s size and the spaces it was being asked to work in. “Each box took maybe 25 or 30 seconds to move.”
We knew there was this other alternative, which is basically Stretch
They needed a more elegant, smaller alternative, which comes in their latest Stretch model. “We knew there was this other alternative, which is basically Stretch,” says Blankespoor “where the arm did a lot more of the motion, and the base only moved when it needed to.”
We pivoted to Stretch about a year ago, really with this early prototype, and then next year, we'll have a product for sale
The change allowed them “to move boxes about five times faster, which, for customers, that's the big driver for return on investment. So, Stretch gave us the advantage of much higher throughput. It's lower cost, much bigger battery life - we can get an eight hour or even a 16 hour battery in the basic Stretch robot - and it's safer. So, we pivoted to Stretch about a year ago, really with this early prototype, and then next year, we'll have a product for sale.”
“There's really three main components for Stretch,” explains Blankespoor as he enthusiastically runs us through their new creation and its capabilities. “There's the arm, there's the base, and then there's this perception mast on the left,” he says, pointing at a perforated metal spar housing wires and componentry sprouting out of the robot’s large, black base, which is approximately the size of a pallet.
They’ve adopted this shape and set of components to try and make a robot that is small enough to work in typical warehouse spaces, agile enough to manoeuvre safely, robust enough to repeatedly handle typical loads, but not so complex as to put it beyond most people’s budgets.
We spent a lot of time making really lightweight joints for Atlas and Spot
Blankespoor emphasises three main components that have helped them deal with the challenges: The robot’s arm, its sensors, and its wheeled base.
All of these have been built on the existing knowledge that Boston Dynamics has built up over the last decade of creating complex, ambulatory robots.
“We spent a lot of time making really lightweight joints for Atlas and Spot,” their dog-like quadrupedal robot, outlines Blankespoor. So “if you zoom in on Stretch’s wrist joints, it's the same as Spot’s hip. They have the same assembly in terms of electric motor, gearbox sensors, even the software that controls that joint.”
We use a lot of the same software to actually understand what all of the cameras are seeing. So, whether it's detecting boxes, or identifying obstacles, that's a big shared code base
Blankespoor claims that this framework means that the arm is around a quarter the weight of an equivalent industrial arm. “That lightweight arm enables us to put it on a much smaller mobile base,” he clarifies.
Similarly, “We use the same depth cameras on Atlas, as we do on Stretch. And, more importantly, we use a lot of the same software to actually understand what all of the cameras are seeing. So, whether it's detecting boxes, or identifying obstacles, that's a big shared code base.”
Blankespoor hopes that they can use their different “technological building blocks” built over the years to “quickly build new robots,” that “can look very different,” but have a common core.
Picking up the pace
However, this doesn’t mean they have everything solved, and there is still plenty to do before the Stretch robot goes into serial production next year.
“The hardware part is important, right, building a robust platform,” says Blankespoor, but he admits that the perception and manipulation of boxes, “is the hard part.” This is due to the robot’s need to operate in “really hard environments,” full of clutter, making software a critical focus.
If you put a box on the floor by itself, it's easy for a robot to find that box, but that's not what warehouses look lik
“If you put a box on the floor by itself, it's easy for a robot to find that box, but that's not what warehouses look like. If you have a stack of lots of different types of boxes that are packed tightly in a truck that you're trying to unload, actually detecting all of those is a lot more challenging, especially when they're right next to each other. It's hard to find the seams between them and then also manipulating them is hard as well.”
This is why they are putting a considerable amount of resources into the software development side. The focus is on “perception and box detection, for control systems and for autonomy,” to make sure the robot can grab “the very top box that’s right up against the ceiling,” using its suction head at the end of its arm.
Every part has been totally redesigned for manufacturability, for reliability, and for higher performance
Doing these things reliably is currently “The critical path for us,” says Blankespoor. However, so is building up for serial production. According to Blankespoor, between their prototype and the ready-to-market version they are finalising now, “every part has been totally redesigned for manufacturability, for reliability, and for higher performance.”
Once Stretch launches and complements their Pick computer vision solution, it will mark a key chapter in the company’s transition from experimenting with robots to patrol alongside soldiers to working within warehouses.
Most of the people we talked to are just really struggling to hire fast enough to support the growth
The past is replete with technologies once funded by US defence spending, but are now deeply embedded in private, civilian companies. Modern computing, GPS, the internet, and voice recognition all have origins in defence projects, as do many more. Will Boston Dynamics’ logistics robots join this illustrious list?
Blankespoor is confident, estimating that there are “500 billion boxes a year,” being moved manually, but “most of the people we talked to are just really struggling to hire fast enough to support the growth that's affiliated with that.” The “people that run distribution centres and fulfilment centres can't keep up in terms of hiring.” If Boston Dynamics can grab just a small slice of that then they stand a very good chance of making the jump.