video

[Segment 1: Introduction to Amazon’s Vulcan Robot]

Amazon’s new robot Vulcan doesn’t look like a person or move like one. This is the robotic gripper. This is the hand. These are paddles that can squeeze the item. But it’s doing a job that until now, only people at Amazon could do, stowing thousands of items into bins.

Our hands are one of the most amazing parts of our bodies. And it’s that physical intelligence, that physical AI that we’re trying to give to the Vulcan robots.

While other robots rely on cameras, Vulcan is Amazon’s first robot that’s able to feel the items it touches, meaning it can handle way more items than past robots. Choosing and placing them accurately.

So on a single shelf, you might be able to create a certain size space. Maybe for a dog toy, but not for a basketball. The robot is doing that assessment.

[Segment 2: Questions About Workforce Impact]

So as Amazon empowers some of its 750,000+ robots with more and more skills like ours, is Vulcan the robot that could finally eliminate the need for Amazon’s human workforce?

There may be a reduction in workforce, but no, the workers will still be needed and the workers will be more efficient.

We went to Spokane, Washington, to get the world’s first look at Vulcan, fully in use on the warehouse floor, and ask its lead creator exactly how it works, why it’s a game changer, and what’s stopping robots from replacing all the people at Amazon warehouses like this one?

[Segment 3: Warehouse Scale and Robot Integration]

So the warehouse is we’re 2.6 million square feet. It has 2,500 full- and part-time employees. Those employees are handling 3.5 million items every week here, with a lot of help from thousands of robots. Amazon says globally, robots help with 75% of all orders.

Amazon is one of the most advanced robot users in the world. Their robot warehouses are an astonishing thing. There’s some 5,000 robotic drive units here ferrying packages around and tall yellow bins filled with items. Instead of a human being going to get the product from the shelf, the shelf comes to the human being.

Except for some drive units, most of Amazon’s robots, including Vulcan, operate behind a fence. To make sure that everyone around them is safe.

[Segment 4: Vulcan’s Development History]

Robotic arms like Cardinal, Robin and the newer Sparrow have been moving items and packages around in Amazon warehouses since 2021, but those primarily rely on cameras for detection and suction for grasp.

So we give the robot a sense of touch using a sensor that measures force in all three directions.

Aaron Parness spent ten years developing robotics for space at NASA’s Jet Propulsion Laboratory before joining Amazon in 2019. Now, he leads the team that developed Vulcan. When the idea first came up, he thought it was impossible.

There was a recent Ph.D. graduate who was trying to get the robot to stow items into the fabric pods. My first impression was, oh, how naive. The real world is going to teach this new grad that robots can’t do these things.

Then the team came up with the gripper with a special sensor and some key movement capabilities. Like sweeping the items to one side. And I had this pivot from, ah, this is never going to work, to, oh my gosh, this is the future.

[Segment 5: How Vulcan Works]

The final result, Vulcan, is a three part system. An employee puts an item from these yellow bins onto a conveyor where it’s picked up by the first part, a moving robotic structure called a gantry. Which is a robot that moves horizontally and vertically.

We’ll take that bag of Skittles and put it into our buffer wall. It waits there until the robot makes a match. So the robot brain is looking for an opportunity where it can create enough space to put that bag of Skittles.

Next, it’s picked up by this robotic round foam disk. We affectionately call that the squash turner. So it reorients the item kind of like it sounds, things like books where you want the titles facing out, or things where you need to align it so that it’ll fit in the space.

Finally, the item is picked up by the gripper. It has conveyor belts built in and so the item feeds forward and backward. The gripper has a sense of force on its grasp axis as well, so we don’t squeeze a bag of Skittles as strongly as we squeeze a physics textbook.

[Segment 6: Vulcan’s Sensing Technology]

That precise amount of pressure and torque is determined by an AI-powered sensor. This is the sense of touch that, until now, experts say hasn’t been put into commercial use in robotics.

The other thing that’s super cool here is this guy. It’s an extended sort of spatula. And the force sensor, the thing that we’re feeling the world, sits right behind here, right on that spatula.

This new ability to adjust pressure means Vulcan can handle up to 75% of the 1 million unique items in inventory here in Spokane. Amazon’s Sparrow robotic arm can handle some 60% of items, while Robin and Cardinal only manipulate packages.

We’re also using generative AI to predict how the items are going to move. So if I push on the t-shirt, is it going to compress? Versus if I push on the bowling ball?

[Segment 7: Vulcan’s Stowing Efficiency]

Vulcan’s final step, finding the perfect spot to stow the item, adds some major efficiency gains for Amazon.

One of the use cases we have for AI is predicting how much space we can make in a bin, and then taking the right item that maximizes the filling of that predicted space.

So a bit like Tetris?

It is a bit like Tetris.

[Segment 8: Humanoid vs. Non-Humanoid Robots]

There is one robot you won’t see all over the warehouse floor. In 2023, Amazon announced Digit, a humanoid robot by Agility Robotics that it’s testing to help organize and move totes. In 2024, CNBC saw digit being tested, but it’s a long way off from being deployed at scale.

These robots don’t need to be limited by our human form. They could be better than us. So why not have wheels? And that’s the question. Why are you putting legs on the thing? Why are you making it humanoid form when there are better shapes?

Vulcan, meanwhile, is not at all humanoid.

Does Vulcan represent a moment where robotics have gone from gimmick to real world application?

To me, it doesn’t matter if the robot has legs or wheels or it’s bolted to the floor. I think the thing that makes the robot useful is having that sense of touch so that it can interact in high contact and high clutter environments. That’s the tipping point for me, and I think we’re right there and it’s so exciting.

[Segment 9: Robot Operating Efficiency]

Amazon says Vulcan operates at about the same speed as a human for now, but it also doesn’t need to rest.

We have down time for maintenance and repairs and those kinds of things, but the robot is able to operate, you know, 20 hours a day.

Which begs the question, why wouldn’t Amazon replace all its workers with robots?

If you build a terribly complicated automated system and it breaks, then everything stops. Taking out the last human is so expensive, it’s so disruptive. It would be a huge investment and an enormous risk.

[Segment 10: Job Creation and Transformation]

Instead, Amazon says Vulcan will create more jobs that are higher skilled, saying it’s spent more than $1.2 billion on free training to upskill 350,000 of its workers in the last five years.

The new jobs are going to be in maintaining and operating the robots, in installing and building the robots. I think it will be an increase in more interesting jobs, and a decrease in the dull, the dirty and dangerous. Humans and robotics will need to be working closely together.

In her 16 months at Amazon, Kari Freitas Hardy has moved from picking to working with the robots, helping when they run into issues.

It would be many decades off to have them just come in and take over. So at this point, it’s more exciting if you ask me to see the growth potential, because that is where it does increase jobs on the backside.

[Segment 11: Worker Training and Compensation]

Freitas Hardy says she’s not making more money in her new role, but Amazon does have a free Mechatronics and Robotics Apprenticeship program, which it says typically leads to pay increases of about 40% in new roles like those at a fulfillment center in Louisiana.

There are ten times more robots than you would find in a building like we’re in today. They have the same 2,500 full-time, part-time workers that we have here, but 30% more of those jobs are high skill technical jobs.

[Segment 12: Vulcan’s Limitations and Safety]

One reason Vulcan can’t replace people is that it only handles items up to 8 pounds, and certain shapes, like round objects, are hard to manage. But it does move fast, which is why it’ll remain behind this fence.

Because we want to go super fast at stowing items. The robots moving quickly and so it is much safer just to keep it behind a fence.

The robotic drive units on the floor with workers are controlled by QR codes and radio signals coming from special vests worn by workers like Freitas Hardy, so they know to slow down and avoid her.

So our pods are moving at quite a little bit of a pace, and you do not want to get run into by a pod. It could drag you around the floor. So again safety is key here.

[Segment 13: Worker Safety Benefits]

In fact, one of Vulcan’s key objectives is to improve worker safety by stowing and picking in the less ergonomic high and low shelves. Limiting workers to mid-height shelves, the so-called power zone, could lower the chance for worker injuries.

Amazon has long struggled with injury rates far higher than those at other warehouses, although Amazon claims those rates have improved significantly.

It’s all-day-long and repetitive motions. I’m up and down the ladder. I’m squatting, I’m lunging. I hurt my knee because I was personally doing some bad body mechanics.

I know that as a picker, if I had an innovation like this where I could have stayed within my power zone, my days would have been just so much easier.

One of our key metrics we look at is safety. So what percentage of the stows is it doing to the top three or four rows, where the employee would have to reach overhead? We have a plan for that metric to get to 100%.

[Segment 14: Future of Warehouse Automation]

The hardest part of doing this job, I would say, is the repetitiveness.

Stower Daniel Di-Maio has been at Amazon more than 11 years.

I expect everything to be fully automated within ten years. I mean, absolutely, the way technology is going. And it has to be because of customer demand, I’m sure.

But at least for now, experts say fully automated warehouses without humans aren’t coming anytime soon.

We used to call them dark warehouses and dark factories. We don’t tend to talk about those anymore. Humans are so flexible. There’s so much that the human can do that now very much we think about, how can I bring in robots without having to make my factory dark?

At one point there were there were rumors of fully dark warehouses that were all, all automated 24 hours a day. Is that still the future?

No, not at all. Like, I don’t believe in 100% automation. If we had to get Vulcan to do 100% of this stows and picks, it would never happen.

[Segment 15: Business Value and Future Deployment]

So if jobs won’t be eliminated, what is the big ROI for robots like Vulcan?

Product returns are incredibly high and product returns are incredibly expensive. Some of them will be because the wrong thing was put in the box. And if you can reduce that, that’s a real cost savings straight away.

For now, Vulcan is only in two warehouses: this one in Spokane and one in Hamburg, Germany, and it’s processed about half a million orders. Existing robotic arms like Robin have sorted about 4 billion packages.

Amazon wouldn’t disclose how much it cost to develop Vulcan, but it says it took three years and a team that’s grown to 250 people. And Parness says Vulcan is a huge business opportunity.

But Vulcan can interact with the world in a more human-like manner, and that gives us a lot more process paths that we can use automation to bring down the cost that our customer pays, and the speed with which we can deliver those products to our customers.