Meeting an Innovator in Industry 5.0 - Subscriber Content

Source: KKK 999/stock.adobe.com; generated with AI

Mouser Electronics recently talked to Erik Nieves, founder of Plus One Robotics. With more than 25 years of experience programming robots, Nieves created Plus One Robotics to be a leading innovator in Industry 5.0—the integration of human and machine workers into the workplace.

 
Erik Nieves  is the Founder and CEO of Plus One Robotics in San Antonio, Texas. With over two decades of experience in the robotics industry, Nieves leads his company under the guiding principle “Robots work. People rule.” His leadership has positioned Plus One Robotics at the forefront of innovative solutions in human/robot collaboration and supervised autonomy, optimizing logistics through the use of high-performance manipulators and advanced vision capabilities.

Mouser: Welcome, Erik. Why does Industry 5.0 need humans as well as robots?
Nieves: Robots are very good at repeatable work. The automotive industry is the largest consumer of robots because it builds thousands of the same thing every year. These capabilities allow you to optimize each robot, making [each of] them responsible for a single specialist role. And that’s very effective. But I was motivated to bring robots into less-structured applications.

What do you mean by less-structured applications?
We work in warehouse automation and logistics. Compared to the industries that have embraced the use of robots, logistics is one of the domains with low repeatability. Instead, systems must cope with much more variation. That’s really the difference—you need variability instead of repeatability. Given the boom in e-commerce, a huge need [exists] for systems that can handle [parcel] variation more quickly and efficiently.

What makes Plus One Robotics different?
Plus One Robotics wanted to address this market, and Industry 5.0 techniques allow us to do meaningful work in warehouse automation. Our systems will perform about 1.2 million picks around the world today. 

To do this, our robots are always considering their next task. To help them decide how to act, we parallelize our sensing and motion. We place a camera above the picking area to image and evaluate all the parcels coming down the conveyor. Edge artificial intelligence (AI) systems then decide which parcel the robot should pick up. By the time the robot has finished dropping off the previous package, the camera has already told it which to fetch next. 

Cycle times in this industry are very important. We are trying to process between 25 and 30 packages per minute. The computer imaging must happen in fractions of a second, but the variation in the shape and size of packages makes automation a challenge. I like to say that the rate of change in the warehouse is greater than AI’s ability to keep up.

Why is that a challenge?
An automated process without embedded sensors to understand its environment is inherently brittle. Most of the time, when an automated system or a robot stops, it’s not broken; it’s confused. This is why we bring human understanding into the mix. 

An automated system performs most of its tasks independently. The robot uses edge computing and an AI model to perform more than 98 percent of its operations. But occasionally, the sensors will image a scene that [the robot] does not understand. This might be a new package type or some other nuance—perhaps a parcel has been opened or damaged. When it identifies an exception, the robot raises its virtual hand for attention. FedEx is a prominent user of our technology; its main hub is in Memphis. We have several robots there, and occasionally, those robots will phone home and ask for guidance. 

A remote human being now gets an alert on a screen. We call these people crew chiefs, and they are responsible for a fleet of robots. When a crew chief gets a notification, it’s an edge case. In any AI system, the most valuable data are the edge cases. We capture those edge cases in real time. This is the reason the company is called Plus One—adding one human being into the control stack makes it so much more effective. The system becomes far more fault-tolerant and less brittle. 

Because the crew chiefs are working remotely, we can identify these exceptions within seconds. The alternative is having someone walk across the factory floor to physically inspect the robot, which might take 10 minutes. In our world, at 25 picks per minute, that means 250 packages are not going to hit the delivery window and are going to be delayed. 

So that’s the difference. While Industry 4.0 is all about the sensors and the data integration, [Industry] 5.0 is about leveraging our people so that the system becomes more effective. The technical name for this is supervised autonomy.

What happens when the crew chief identifies the exception?
The exception is logged. The machine should only be handling that exception once; it will learn how to automate the situation when it occurs again. The number of exceptions should gradually diminish over time, but there will always be exceptions. 

We’re always introducing new package types. The first time you ordered a book, it came in a cardboard box. The next year, when you ordered the sequel, it came in a carton wrapper. Now, it would be shipped in a bubble pack—we will never cease to innovate. And we learned something else: advertising. You might reorder the same item next week, but now it’s in the Valentine’s Day promotional packaging. Each of these exceptions needs to be relearned.

Does the machine only need to see the exception once?
In practice, it’s more than once. The crew chief might need to identify the edge case five times, perhaps more. Eventually, the AI network says, “Ah, I see a pattern here.” 

After all, AI excels at patterns, but you can’t build a pattern from one data point. The only agent that can handle exceptions on a true one-off basis is a human.

What are the barriers to adoption?
Our domain is a very reticent adopter of technology. A warehouse might have been in operation for 40 years, and its operations manager has been doing this job for 25 years. A reluctance to be told that there is a better way to work [exists]. 

Younger generations are much more comfortable with robotics. The social acceptance of robotics will improve, which will ease their adoption into industry. This is where [an Industry] 5.0 mindset is so helpful—its philosophy is that we’re going to make your associate’s life better and improve your throughput. 

We also face the challenge of data security. All Plus One robots share a single AI model. Think of it as one giant set of experiences that has been accumulated over millions of individual operations. Some customers are uncomfortable with this big AI model—they believe that their world is unique and that their data should only be for them. 

If that’s what you want, we can accommodate you. But we will have to repeat the learning experience all over again from scratch—there really is no need [to do that]. Consider a single package: If you ship it through FedEx, we will see the same package again down the chain. If FedEx isn’t delivering the last mile, it’s being handed over to the U.S. Postal Service, and we will see it once again. The AI model doesn’t care. It just needs the sum of its experiences—and that of the crew chiefs—to be truly effective.

So, placing the human in the middle of the process is the most important aspect of Plus One’s innovation?
Yes. More robots will be in these domains all the time, and not enough labor is left to continue growth. Information can only get you so far, and sometimes you need intuition and nuance. We’ve built our processes around our human capabilities, and we have an immeasurable capacity for the unique. That’s what makes Industry 5.0 so powerful. Making the human a key part of the automation makes it far more effective than the sum of its parts.