BOSTON — When it comes to using robotics in the warehouse, the one thing everyone can agree on is that physical AI promises to enable autonomous machines to perceive, understand, and adapt to changing conditions in the real world.
However, the actual presence of physical AI is not yet as prevalent.
“There’s been a lot of progress in robotics in the last few years, and ‘physical AI’ is now a very popular term,” said Omar Asali, chairman and CEO of Ranpak, a packaging systems provider.
“I think the biggest issue that we see is that, frankly, deployment at scale and mass adoption have not occurred,” he added. “I think the promise looks great. There’s a lot of beta testing and alpha testing, there’s a lot of demos, but real mass adoption has not occurred yet.”
Asali outlined the progress and challenges in a session at last week’s Robotics Summit & Expo on how physical AI is driving profitable and scalable sustainability in warehouses. He spoke about a range of issues with Sarah Wynn, senior editor of Packaging OEM (a sibling site to Automated Warehouse). They included the need for better data coordination, the potential of AI in predictive maintenance, as well as a future where more automation drives tangible return on investment (ROI) and efficient use of resources.
Mobility versus manipulation
Ranpak uses large robotic arms for palletizing and de-palletizing. In addition, Asali said one of the investments that the OEM has made is in Pickle Robot Co., which automates package loading and unloading from trucks and trailers.
Pickle Robot uses pre-trained models to create generative AI applications specific to the logistics domain that can work at human — or even better-than-human — speed.
While many deployments of warehouse automation include autonomous mobile robots (AMRs) that transport goods from one point to another, the ability to manipulate goods versus move them is an important distinction, observed Asali.
“The piece around mobility is more advanced than the piece around manipulation, so I think there’s a little bit more scale there, and you see that in deployments,” he said. “I think the manipulation piece has been harder.”
Asali noted that the biggest reason why robot-enabled manipulation hasn’t been deployed is twofold: First, a lot of companies have claimed that certain capabilities would generate ROI that has not materialized, which impacts future deployments.
Second, the issue is not that robots cannot do something, because they can do a lot, added Asali. The issue is doing it in a way that generates the right payback for customers. “I think that’s the question that hasn’t been answered satisfactorily,” he said.
Also, many of these robots are task-specific deployments that are less likely to scale. “But you have to start somewhere, and eventually these application-specific robots will evolve to more multitasking systems,” Asali acknowledged.
“This is no different than when Amazon started as a book business and became an e-commerce machine,” he said. “But they needed to do the book business first to perfect billing, logistics, and the customer experience and then expand.”
Despite the progress, Asali said there remains a gap between technology capabilities and what has been deployed at scale. “The pace of innovation versus the pace of deployment — there’s a disconnect,” he said.
Physical AI makes data-based decisions
In response to his Amazon analogy, Wynn asked Asali for practical examples of how robots can help bridge the ROI gap and even add value. The short answer: data.
Ranpak’s DecisionTower is a vision system that uses AI to detect out-of-scope boxes to optimize packing and prevent machine stops, thereby helping to maintain equipment uptime. For example, DecisionTower can look at the flow of boxes coming in, and if there’s an overflow or issue of some sort, the vision system will indicate that the parcels need to be diverted rather than move to the equipment.
In the future, companies will focus on better orchestration and coordination of the data they have in different systems, Asali asserted.
“We work with some of the most sophisticated companies in the world, and we get their data from their warehouse management system [WMS], and what’s shocking is how poor that data is, how inaccurate it is,” he said. “So even today, as we talk about AI, we live in a world where the data that we’re all using still can be improved significantly.”
“I think over time, as we automate more of these tasks, I think we’ll see quite a bit of improvement in data integrity and accuracy, which will be self-fulfilling in making robots even more effective,” said Asali.
To help solve some of these problems, Ranpak has been using AI to make its own equipment better. To that end, the machine will respond to the operator through the company’s Connect.IQ suite. Workers could literally talk to the equipment and then get feedback from the machine about potential problems and how to service it.
As a result, a machine operator can spend less time talking to Ranpak technicians and more time solving problems, said Asali.
“We’re doing a lot of data-analysis work for preventive maintenance and predictive maintenance,” he said, noting that it will not only help with equipment uptime, but also with overall productivity. “The equipment is just becoming smarter in an industrial setting to bring efficiency to the customer,” Asali said.
Eventually, using physical AI, that efficiency could scale across the warehouse.
