Walmart has been spending aggressively on tech to close the gap with rival Amazon and bring new efficiencies into store operations. But the legacy retailer has found that similarly legacy databases and raw, unstructured data can stymie efforts to forge ahead in the artificial intelligence arena.
“At Walmart, we’ve been working really hard to…see how much technology we need to add to our stores to really drive that benefit back,” Ryan Kee, Walmart’s director of innovation and AI, said Wednesday at the AI Summit in New York City.
As the catch-all umbrella term for numerous technologies, AI can be intimidating and overwhelming for retailers trying to figure out where to get started. But Jason Nichols, who recently left his role as director of AI in Walmart’s Intelligent Retail Lab to head up data science and engineering at Centricity Insights, says robotic process automation (RPA) is where retailers can unlock the greatest value. Over the next three to seven years, RPA holds significant potential for retail store operations, Nichols added.
“At its heart, one of the big problems that faces AI in any [physical] environment is an absence of agency,” Nichols said, alluding to the camera-equipped robots that traverse Walmart’s stores monitoring inventory and taking on the task tidying the chain’s well-trafficked floors. “A lot of these systems are easily baffled by small environmental factors.”
One of these environmental factors? People.
Shoppers going about their business and browsing store aisles occlude “30 percent to 40 percent” of the training data gathered by roving bots that gets fed into retailers’ computer vision systems, rendering it of limited use, he added. “It’s really easy for a human to take a half step over and get a better view of the problem,” Nichols said of robots’ limited ability to self-correct and pivot on the fly. “It’s really, really hard for fixed, mounted cameras to do that.”
Related Stories
The next generation of these kinds of robots will need more advanced algorithms, software and cameras that can pan, tilt and zoom in order to work around these current limitations and generate actionable data and insights.
In April, Walmart deployed 1,500 automated floor cleaners the company said would take over the dreary, repetitive task from store employees. The Bentonville, Ark.-based retailer also unleashed hundreds of shelf-scanning bots across stores that, beyond keeping tabs on product stock levels, ensure items are in the correct location and advertise accurate prices.
It’s all too common for robots confounded and overwhelmed by their surroundings, Nichols said, to end up stuck in a corner after going out of their way to avoid bumping into a customer. The sight of a towering, 800-pound robot stranded “helplessly” is “pretty pathetic,” he added. Bots that find themselves in such a predicament require someone to physically move them to a new location, limiting the benefits of their so-called automation.
With its size, reach and scale, Walmart has been assuming first-mover status with leading-edge robotic technology. But until RPA evolves to more effectively and intelligently operate in and add value to the store environment, it’s entirely possible robots will remain out of reach for retailers lacking Walmart’s deep pockets.