Skip to main content
science at work

Huge retail warehouse market seen for the Sort system, made by Canada’s Kindred Systems Inc. The man-machine hybrid features a robot arm and a grasping technology powered by reinforcement learning and, yes, that means it learns from humans

The Sort system is a man-machine hybrid that incorporates a robot arm and a grasping technology that is powered by reinforcement learning.Jeremy Francis

When Jim Liefer was vice-president of operations for Walmart’s online business, his biggest problem was finding workers to sort customer orders in the company’s warehouses. There just weren’t enough of them.

Robots should have been the solution, but there was just one issue.

“Robots have been out there for a while in manufacturing – and they’re pretty dumb,” he says. “Dumb robots can only do so much. People need to have smart robots.”

That simple realization is ultimately what led Mr. Liefer to join Kindred Systems Inc. as chief operating officer in late 2016, after a stint at the luxury furniture company One Kings Lane. Kindred, founded two years prior and co-headquartered in Toronto and San Francisco, was on a mission to add intelligence to robots.

Two years later, Mr. Liefer says the company is succeeding in that goal – making robots potentially more useful to large retailers. A number of them are now using Kindred’s Sort robotic arm, a system that can sift through products in a large bin, identify them, then arrange and organize them for further processing.

This smart robotic system, he believes, is going to be vital for companies as they further integrate e-commerce into their businesses.

“They’re all fighting for the same resource, which is the human. They can’t find the people,” says Mr. Liefer, who took over as Kindred’s chief executive officer and president this year. “All of these retailers have woken up to the fact that the biggest existential threat to their business is they just can’t find people to do the job.”

Jim Liefer, Kindred System Inc.’s chief executive officer and president.

The Sort system is something of a man-machine hybrid that incorporates a robot arm and a grasping technology that is powered by reinforcement learning.

The arm uses sensors to identify products in a sorting bin. Kindred’s “AutoGrasp” algorithm then determines the best grip angles and pressure for the arm to use.

It also calculates the proper movement path to take so that the arm doesn’t hit other items and drop what it’s holding. Finally, the algorithm tells the robot where to place its cargo.

But, because retail products come in various shapes, sizes and forms, the Sort robot does often encounter situations in which it doesn’t know what to do. In such cases, it calls in a human operator in Toronto, employed by Kindred, who can take over remotely and execute the proper sorting.

Kindred’s secret sauce lies in its algorithm, which learns as it goes. The Sort robot gets better by observing the human operator’s actions.

Mr. Liefer says that in initial tests last year, the system successfully grasped objects only about 20 per cent of the time. Since then, it has improved to about 80 per cent.

“Each time it executes its task, whether it’s successful or not, it learns from that,” he says. “Our human is being piped in only 20 per cent of the time.”

Kindred’s secret sauce lies in its algorithm, which learns as it goes. The Sort robot gets better by observing the human operator’s actions.Jeremy Francis

Mr. Liefer believes Kindred can get that result up to 90 or 95 per cent and that it can save retailers money with its robots-as-a-service model, where they essentially subscribe to Sort and pay for each individual item placement.

The company has attracted several high-profile brands as customers as a result, although the only one Mr. Liefer will name – citing competitive reasons – is Gap Inc. The San Francisco-based retailer deployed two Sort robots last year at a fulfilment centre in Gallatin, Tenn., just outside Nashville, and now has six in operation there. The company has also added an unspecified number of robots at its operation in Fresno, Calif.

“We saw enough potential in the prototype that we wanted to give it a shot,” says Kevin Kuntz, senior vice-president of global logistics fulfilment at Gap. “It’s an accurate system, on par with our hourly workers. It’s not any faster, it’s not any slower, at least today, but we see a lot of potential for it getting faster in the future.”

That potential has drawn interest from investors. Kindred last year announced a $35.4-million Series B round of funding led by China’s Tencent, with participation from Eclipse Ventures and First Round Capital. The round brought Kindred’s total announced funding to US$44-million.

If the Sort robot encounters a situation in which it doesn’t know what to do, it calls in a Kindred human operator, who can take over remotely.Jeremy Francis

The company has used the funding to grow to 80 employees, split roughly between offices in Toronto and San Francisco. Two of the company’s original founders, Geordie Rose and Suzanne Gildert, have stepped away to focus on Sanctuary Cognitive Systems Corp., a spinoff that is aiming to make robots with human-like intelligence.

Industry observers believe Kindred’s pivot toward servicing the e-commerce market is a smart move.

“There’s all sorts of potential to having a system which can operate with the same reliability and flexibility as a human. There’s a huge market there,” says Ryan Gariepy, chief technology officer at Kitchener, Ont.-based Clearpath Robotics Inc.

“The question which really comes up with e-commerce and these other markets is: dollars matter. Speed is money, if they’re able to operate fast enough to make it worth it for these companies.”

Other observers agree that the potential market is big, but there are unanswered questions about whether the technology is ready to operate at the speeds necessary to make it widely valuable.

“It’s a very uncertain environment, it’s very dynamic,” says William Melek, director of mechatronics at the University of Waterloo. “It’s tough to estimate how long it will take them to get there, whether it’s a matter of a year or two or five.”

Will robots take over jobs?

For much of the past few years, the prevailing narrative about the rapid advancement of robots and artificial intelligence has centred on the possibility that they will bring about mass unemployment.

A landmark 2013 study from the University of Oxford provided much of the ammunition for that line of thought, predicting that nearly half of all jobs in the United States were at risk of being automated over the next 20 years. With machines getting better and smarter, critics argued, there is a growing possibility that humans won’t be left with much to do.

Some newer studies are countering these predictions. A PricewaterhouseCoopers report released this summer, for one, suggests that robots and AI will actually create more jobs in Britain than they will destroy over the next two decades as they open up new markets and human occupations.

“In absolute terms, around seven million existing jobs could be displaced, but around 7.2 million could be created, giving the U.K. a small net jobs boost of around 0.2 million,” the report said.

The same is likely to happen in other developed countries, experts say, as advanced automation becomes more complementary to human workers – as Kindred’s Sort system is – rather than a replacement for them.

“People are beginning to realize that robots will not be able to achieve everything even with very advanced AI,” says William Melek, director of mechatronics at the University of Waterloo. “It’s going to be more of a hybrid environment.”