Why Siemens Is Building Collaborative Robot Workers

Two minds are better than one. JL

Sean Captain reports in Fast Company:

Instead of doing one job like typical factory bots, Siemens wants autonomous robots to work together to figure out how to tackle projects. A swarm of general-purpose worker robots that can be assigned a task and then figure out among themselves how to get it done.
In the 1936 film Modern Times, Charlie Chaplin plays a factory worker whose only job is to tighten two bolts—again and again, all day, until he finally goes mad. It's the life of his robot descendants, which might, for example, weld the same car part over and over again. That model is reaching its own breaking point, says German industrial giant Siemens, because it's too clunky to keep up with market demands.
"We're going to see more complex products that consumers or different industries want us to manufacture," says Livio Dalloro, head of research for Siemens Corporate Technology. "The costs to bring [an assembly line] up and then essentially bring this down, when you are going to be switching to a different product, are pretty high." In other words, the costs of reconfiguring a traditional production line for a new product get in the way of being able to quickly iterate on product design.
Siemen's solution: a swarm of general-purpose worker robots that can be assigned a task and then figure out among themselves how to get it done. Dalloro's team at Siemens's lab in Princeton, New Jersey, has built spider-like 3-D printers that crawl around as a coordinated team. Looking like an arachnid Wall-E wearing a fez, each bot has a head with a 3-D depth-sensing camera similar to an Xbox Kinect. On top of that is an infrared laser scanner to further gauge the surroundings. Off-the-shelf circuit boards running Linux on multicore CPUs handle tasks like analyzing the immediate terrain. The spiders chat over Wi-Fi and Bluetooth, reporting how much ground each can cover so that their collective mind can break up the job and assign out portions of it.
Dalloro gives an example of how the system, called SISPIs (Siemens Spiders), might eventually tackle a project. "This part essentially comes out of a . . . 3-D drawing, and this gets communicated to the robots," he says. "And the robots essentially receive all the same information, and they negotiate with each other . . . who will manufacture each part and how to split the execution . . . in a parallel fashion."
The spiders know their capabilities and limitations. Each is fitted with three gyroscopes and accelerometers, plus actuators in the legs that measure force—all in order to determine the spider's position and how it is moving from one spot to the next. The team can even figure out how to cover for a robot if it breaks down or its battery dies.

The Hive Mind Factory

Siemens' ultimate goal is not to deploy an army of plastic-spraying 3-D-printing robots. SISPIs are a proof of concept demonstrating that a flexible, autonomous robot team is possible—an idea first proposed by Siemens engineer Sinan Bank in 2014. "We wanted to do it as quick and dirty and fast as possible," says Dalloro. In a video showing one of the robots in action, the bot stomps around tentatively like a newborn foal still finding its (six) legs. "This is the first step for us," Dalloro says, calling the overall project a moonshot. "This allows us to accelerate and develop a variety of different technologies in multiple disciplines at the same time," he says.

"This is the first step for us."

One challenge, obvious from the prototype, is to make the robots more graceful. Bank and Dalloro will use machine learning to analyze how the robots move and figure out better ways to do it. Not all future robots might move like spiders.
Siemens is also figuring out how the robots should navigate. These bots currently use a visual method called PTAM (parallel tracking and mapping) that allows them to recognize and navigate a setting without requiring special markers to orient themselves. (The technology has also been used in augmented reality to map a landscape and superimpose virtual objects like cartoon characters on it.) In the future, the robots may also triangulate using radio beacons.

Related: How Soon Before Your Job Is Done By A Robot? This Handy Chart Will Tell You

Though the current robots are about the size of a microwave oven, the basic design can be scaled up to industrial size, says Dalloro. Robots that weld cars together one day, for instance, might reconfigure themselves later to weld steel cabinets. "For sure, we can see the applicability for a reconfigurable factory [with] more general-purpose manufacturing machines that can move around in an autonomous way," he says, "rather than very highly specialized and expensive space-confined machines."

The robots might not even be confined to a factory. Dalloro says that a contingent might go out to a harsh environment and build a base for future settlers by printing with concrete instead of plastic. Continuing the moonshot metaphor, I ask him if this base might be on another world. "That would be even more exciting," he says with a laugh, adding that the scenario is way beyond what he's currently envisioning (though not out of the question).

Dalloro and Bank aren't really sure exactly where the technology will go, and they don't see that as a problem. SISPIs grew out of the Siemens Seed Fund—a kind of skunkworks, established in 2012, that funds ideas by its scientists and engineers that are not connected to any products the company is currently working on. Dalloro aspirationally likens it to the U.S. military's Defense Advanced Research Projects Agency (DARPA). A better comparison might be a mini version of other free-ranging company labs such as Microsoft Research.

"I don't know what the next version will look like," says Dalloro. "But I can tell you that there will be a next version. That's for sure."