Feature Story

Startups tackle cloud robotics

by Anne Field

Startups tackle cloud robotics

Tech startups are tapping the capabilities of cloud robotics.

Thanks to such advances as sophisticated mobile communication technologies, more robotics applications are happening in the cloud. Called cloud robotics, it allows robots to take advantage of the mega-computational, storage and communications capabilities of data centers. And it also ties robotics to the Internet of Things, big data analysis, and sensor networks, to name a few areas.

As you'd expect, a growing number of tech startups are tapping into these capabilities. Here's a look at three such companies:

How does this grab you?

As ecommerce growth explodes, warehouses are struggling to keep up with demand and fulfill orders quickly and efficiently. With that in mind, an increasing number are introducing robots into the mix. But there's a big problem: the ability to grip. Specifically, robots typically have trouble grabbing small quantities of many different-sized items. An aspirin bottle, for example, doesn't share the same weight, size or shape as, say, a pair of socks or a teddy bear. And those disparities have been a tough nut to crack for robotics experts.

See also: The Cisco and Google Cloud partnership

That's where Cambridge, Mass-based RightHand Robotics comes in. Its RightPick system addresses the tricky task of grabbing individual items--known as piece-picking--and allows robot to assess the shape, size and durability of thousands of different items. The key ingredients include a machine learning backend and robotic hands able to sense whether an object needs to be handled gently; those hands work with existing robotic arms, so they don't require a lot of new hardware.

"Our mission is to enable robots to grasp and manipulate everyday objects," says co-founder Yaro Tenzer.

The system takes in information both visually and with tactile sensors, then storing that data in the cloud. As a result, robots can share data with their brethren on the warehouse floor and learn from experience, a capability Tenzer calls "fleet learning." "Every time they pick something up and put it down, they collect data and it's sent to the cloud," he says. "They keep learning and getting better and better." The company integrates the RightPick system into a warehouse's control and management systems, which tell robots what to grab, where and when.

Co-founded in 2016 by a team of researchers from Harvard Biorobotics Lab, the Yale Grab Lab and MIT, the product just came out of stealth mode last spring.

Super-fast decision-making in the cloud

Soodeh Farokhi has a particular interest in the matter of latency—for example, how quickly decisions can be made through cloud computing. And that led her to found her startup C2RO Cloud Robotics, which has a robotic platform through which subscribers can tap a variety of services, ranging from facial recognition to obstacle detection.

It started a few years ago, while Farokhi was getting her PhD in technical computer science at the Vienna University of Technology, and she and her husband decided to relocate to Canada. They ended up moving to Montreal, where Farokhi signed on at the TandemLaunch incubator, which houses experts working with emerging technologies and gives them $800,000 in seed money. Eventually, that led to her working with researchers at the University of Indiana and developing the technology that formed the basis of her startup.

See also: Cisco innovates in fog comupting

The crucial element, according to Alex Gaughan, head of SAS marketing, is parallel processing, one of the company's three patent-pending technologies. That makes it possible to process applications and large amounts of data in the cloud, while also doing some processing locally on a robot--without any lag time between the robot taking in information and a cloud application interpreting it. That allows, say, a robot to take snapshots of hundreds of faces and send them to a cloud platform, where facial recognition analysis can be done quickly and efficiently.

Customers sign up for a monthly subscription, which potentially gives them access to multiple services. Perhaps the most popular, according to Gaughan, is C-SLAM. (That stands for Simultaneously Localization and Mapping). It allows robots to navigate complex and unstructured environments without banging into anything. (Think a vacuum cleaner able to figure out the fastest route from the living room to the kitchen and then easily follow that path).   

Still in beta, the product should be commercially available later in 2018, according to Gaughan.

Monitoring robot performance

Robots in the production process add efficiency, lower costs, speed and new capabilities, among other features. But if your plant depends on a lot of different robots, it's pretty important to know how well the systems are working and whether there are any problems—and to do so remotely. And you want to avoid adding a lot of extra equipment and going through a complicated integration process. 

Startup Tend's in.view software targets that market. It allows you to monitor and analyze robot performance remotely and in real time through a cloud-based application on a mobile device. You get automated alert notifications, as well as live and historical video feeds and other data, so you can drill down and find out more detail about whatever is going on with the robot in question.

The software is an outgrowth of an original cloud-based platform allowing users to control multiple robots from a single device remotely, in addition to monitoring them. But after customers demonstrated particular interest in the monitoring and analytical capabilities of the program, "We commercialized that as a separate product," says Eric Foellmer, chief marketing officer of the Boston-based company.

Customers can customize their alerts or they can arrange for different notifications to be sent to various employees, depending on their role. A production manager who's off-site, for example, might want to know whether a piece of equipment has stopped working, not when it's likely to break down next. "They don't have to get a lot of notifications that aren't important to them," says Foellmer.