In 2018, the Amazonteam began building a new backend system to maintain their vast and varied fleet of devices. With the new Comprehensive Device Management (CDM) system, multiple teams can deploy new robots, update the devices with minimal operational disruption and manage security across fleets. The centralized management layer was built with Amazon Web Services IoT.
Now that Amazon is rolling out CDM across its facilities, Amazon CTO Werner Vogels has detailed the system in a new blog post. CDM is "the heart of what makes Amazon's robotic operations truly smart," he wrote, "a leap in evolution from robotic devices of the past that rely on consistent conditions."
Speaking with ZDNet, Vogels elaborated on what it takes to build a "truly smart" robot.
"I think it starts with being able to collect data that you have, that understands what's happening," he said. "Just having the data about, let's say, the whole production line that you're trying to introduce robots into."
With the right data, Vogels said, manufacturers can forgo common but potentially costly practices, such as "stopping the line" when a defect is spotted.
"If you have to rebuild your robot because you have the next generation of your car coming, that's a very expensive operation," he said. But if robots are more configurable and more adaptable to some of the software that you load onto them, they have more capabilities. You can also immediately address a defect that is happening. And I think that is a very important part of modern robotics, having them be adaptive."
A program like CDM, he said, showcases how manufacturers -- like Amazon's retail arm -- can leverage AWS IoT to build a whole fleet of devices that are adaptable.
"What we're focusing on, at both AWS as well as at Amazon, is to try and build these robots that can actually be reconfigured or that can be adaptive to new tasks, especially because we're continually innovating and building new capabilities for these robots," he said. "So you need to continue to update the software that actually runs on the robot itself. And you need to be able to do it not just for one, but for 40,000 or 50,000. And given the huge variety of robots that we're using, it's a big challenge."
CDM, Vogels pointed out, also demonstrates that a "better robot" doesn't necessarily look dramatically different from its predecessors.
"The whole digital-physical world is massively changing, and it's not just in our fulfillment centers or in manufacturing environments," he said. "If you look at the new Audi, for example, I think it has 1,200 sensors in it. If you have a Tesla, you understand that this is a computer on wheels -- but the new Audi looks just like the old Audis. They're not fancier; they look just as conservative. However, they are computers on wheels, and people that actually use these Audis don't think they're sitting in something that is actually just as advanced as Telas because the magic happens behind the scenes."