Long used to handling boxes and some standard shaped items, picking robots are on the verge of moving into a world where they have to distinguish between thousands of quite different items. And they have to make picking decisions faster than people. Getting there will not be in the blink of an eye.
In real estate, it’s all about location, especially if it’s a distribution center location. But inside these DCs, it’s rapidly becoming all about labor. Or lack of labor, actually.
That convergence of location and labor in distribution are rapidly making picking robots a newfound possibility out on the floor. “Picking robots are a straight labor availability play,” explains Erik Nieves, CEO and co-founder of PlusOne Robotics.
He goes on to explain that DCs in many locations are clustered. “And when you burn up labor in a 40-mile radius, no one is moving there to become an order picker. This isn’t Detroit in the ‘50s and ‘60s when people moved there from all over the country to build cars.”
Fortunately, picking robots are on the cusp of their next big move in DCs.
In fact, picking robots have been on the floor for some time from palletizers to picking certain individual items such as pill bottles and boxes. But that’s a very limited range, especially in a world of e-commerce with thousands upon thousands of SKUs.
“So many e-commerce companies including retailers doing click-and-collect at physical stores have recently become highly receptive to automation including picking robots,” explains Vince Martinelli, head of product and marketing at RightHand Robotics. “The early adopters want to get ahead on this given the labor challenges they already experience,” he adds.
“The idea is to reduce labor intensity, especially in e-commerce sites, and reduce click to ship times,” explains George Babu, co-founder of Kindred.
This month Modern Materials Handling did a cover story on Gap Inc.’s use of Kindred picking robots to feed a put wall at its Gallatin, TN e-commerce DC. While this application of picking robots is only a small portion of a half million square foot DC, it is a notable one.
To begin, it is an important example of Gap Inc.’s embrace of automation. “In the past, distribution was viewed as a cost center and now, because of innovation, it’s a competitive advantage for us,” says Kevin Kuntz, senior vice president of global logistics.
Now more than a year after an initial pilot, Gap Inc. uses picking robots at six locations in the facility. In addition, the equipment has been introduced to the company’s Fresno DC.
So, what makes picking robots so attractive? Martinelli explains it this way.
“On the robot side you start with the 3Rs – range of items, rate of pick and reliability. On the customer side, the outcome is higher throughput, utilization and quality,” he says.
And recent advances in robot-related technology are adding an exponent to that equation. “Furthermore, we are riding a tech wave that wasn’t cost effective just shy of five years ago,” says Martinelli.
Key components here are robotic arms, end effectors/grippers, vision systems, software and the star of the future – artificial intelligence. AI is so important because it will bridge picking robots from the highly structured environment they live in today to a highly unstructured world, especially in e-commerce.
“The future of picking robots is handling thousands of items that are not much alike,” explains Dematic’s Crystal Parrott in the NextGen interview this month. Furthermore, “the robot does not know in advance what the next item is and has to adapt to it on the fly. Welcome to the unstructured world,” she adds.
As Nieves explains, “we are entering the early stages of reinforcement learning in AI. This is a capability that hold the keys to the kingdom for AI. We want robots to learn as quickly as possible how to handle thousands of different SKUs autonomously.”
As Babu explains, “we are finally at a point where intelligence allows robots to pick items. Reinforcement learning is the backbone of that. And today it is emerging, showing up in applications on DC floors.”
But these are early stages for reinforcement learning. “We’re still quite some distance from full autonomy across the board,” explains Babu.
One of the great obstacles here is the simple matter of exceptions, says Martinelli. In fact, exceptions are the most unstructured portion of the unstructured world.
This is a big part of the reliability that Martinelli referred to earlier. It’s one thing to pick the right item consistently. It’s something else entirely for a robot to know what do if it drops an item or the wrong item shows up. “How robots deal with exceptions is existential to the robot. Not my fault, man, doesn’t cut it,” says Martinelli.
But as Babu points out, “variety of all kinds makes for better training data for the robots.” In fact, exception handling is enhanced with assistance from humans in the loop.
Clearly, we are not at a point where picking robots are ready to entirely make a go of it without any people. That’s okay, says Nieves. In his view, robots are better off when teamed with people. “Several robots teamed with a person is much more effective than just robots,” he says.
And it’s likely to be that way for some time to come.
Gary Forger is the special projects editor for Supply Chain Management Review. He can be reached at firstname.lastname@example.org.