Identifying the Problem that AI Can Solve with Rapid Robotics' Juan Aparicio

In the third of our series of A3 interviews with AI leaders, Juan Aparicio, VP of Product at Rapid Robotics, discusses AI use cases in manufacturing, with an emphasis on robotic manipulation. Aparicio, who sits on A3's Artificial Intelligence Technology Strategy Board, recommends identifying and understanding the problem before jumping into discussions around which tool or technology to use. Check out his advice on using robotics-as-a-service and the right partner to start automating in your facility. And, you can get more of Juan's advice on September 29 in Columbus, Ohio at the AI & Smart Automation Conference.

Juan Aparicio

Juan Aparico
VP of Product
Rapid Robotics

How would you advise companies to choose their artificial intelligence projects – and what questions do they need to answer before they begin?
Quoting Bill Gates, applying automation to an efficient operation will magnify the efficiency, but if you apply it to an inefficient operation, it will magnify the inefficiency. So my first piece of advice is to understand what tasks you are trying to automate, either because you have labor shortages or you want to tap into the other benefits of robotics and automation, and see if AI or even automation is right for it. Humans are extremely dexterous and adaptable, robots not that much. However, robots bring repeatability, 24/7 operation and higher speed. If you determine that the task you have in mind is not the right one, look around, most likely there are other pick-and-place operations being done by humans that are ripe for automation. 

My second piece of advice is that, if there are different candidate tasks for automation, start with the low hanging fruits and get some early wins. I am not talking about free pilots and proof of concepts. Start small and have a plan to scale, but get an impact from day one. 

If the task is a good fit for automation, my third piece of advice is that, before you go into the AI route, look for proven automation technologies and see if they may be adequate. Ask yourself what value is AI really bringing. Try to simplify as much as possible. For example, can you modify the part presentation so a camera running pose estimation is not needed? You will save in costs, and gain in reliability and speed. 

Then, if you determine that AI is the right way to go because of the promised flexibility, my next advice is to be sure to understand the risks, the maturity of the technology and the entire infrastructure needed to support it when tackling it on your own (training data, edge/cloud, connectivity, engineers,etc). Keep in mind that it is not unusual for key information in manufacturing to be kept in paper form. Having your “data house” in order is a prerequisite for data-centric automation techniques. AI implementations will also require a close IT/OT collaboration, and that’s usually easier said than done, as their goals may not be aligned. 

In summary, don't choose the tool before you understand the problem. Find a partner to go with you along the process. A partner that is committed to the solution and not to a specific technology. The AI skills gap is often cited as a roadblock to AI implementation. The competition for computer vision, data scientist and other AI domains is fierce. So look for partners and vendors that are laser focused on solving your labor issues and are able to bring AI when and where is needed. 

How much talk about AI right now if hype vs. reality? Where is AI having the most impact now in manufacturing and automation? What are some of the effective real-work use-cases for AI that are being deployed today?

In every deployment of AI that is in production right now, either there is a human in the loop, or the risk for failure is very low. If the output of the AI algorithm can have a catastrophic outcome if it fails, then it may not be the right target application. For example, missing a pick in a bin picking operation, if you maintain the overall throughput, is not a big deal. That is the reason why you see many companies addressing bin picking, palletizing, kiting, etc. Problems that involve flexible grasping are now achievable. But there is a big caveat, the technology is object specific, and your specific parts may not be so easy to grasp! In addition, a lot of emphasis has been put on picking, due to the high demand in logistics. But precise placing is also essential in many manufacturing applications and machine tending. 

Quality inspection is also an effective use case. Using AI and machine learning for inspections and defect analysis is one of the most common applications of AI in use in automation today. In a recent survey of Association for Advancing Automation members, 82% of companies that had implemented AI technologies said they were using vision/inspection applications in their processes.

Another target area for AI are autonomous mobile robots (AMRs). AMRs are already leveraging AI to make better decisions on routing, optimizing path efficiency and learning how to navigate better around humans, and to differentiate relevant objects in the environment like forklifts.

On the frontier line of manufacturing applications soon to be impacted by AI, assembly is a big candidate. If throughput is not a problem, visual servoing and deep reinforcement learning (accelerated via simulation) can tackle relatively complex assembly tasks like peg-in-the-hole insertions. But once more, you are paying a price, in this case throughput and cost (force torque sensors, cameras, computing devices). There is also an interesting area of research around using AI for robot programming assistance, providing suggestions and corrections during robot task programming to help novice users. An example is the project funded by the ARM Institute with READY Robotics, Fedex, Univ. Memphis and DeepHow as partners.

Can you share an AI or smart automation success story with us?

AI and smart automation success stories are difficult to find because when a technology implementation goes bad, the company doesn't speak about it, and when it goes well, they keep it as a secret competitive advantage.

Deployments such as the one from Plus One Robotics at Fedex, or Ambi Robotics and Pitney Bowes or Knapp and Covariant are good examples of AI success stories in intra-logistics setups. 

On the manufacturing side, I can share a technology we recently unveiled, which is already making a tremendous impact at our customer sites.It is an example of computer vision technology that we have developed at Rapid Robotics that enables SMEs to automate in high-mix environments, with the right ROI. It is called Smart Setup and unlocks a capability that until now was out of the realm of automation: easily and quickly move robots between tasks.

The variability in most of SME business is just too high to go all-in with a fixed and costly automation solution. Smart Setup is an example of a 0 to 1 unlocking technology. Without easy task switching, the ROI is simply not there even for simple tasks. SMEs cannot pay system integrators to come in every time they want a robot working on a different task. They'd lose more money than they'd save with automation. And that is the status quo and part of the reason why the majority of manufacturers in the US cannot afford robotic technology. 

This is how it works. With one touch in a tablet-based app, the Rapid Machine Operator (RMO) cameras analyze the workspace. With another, the RMO calculates distances and orientations of parts and surfaces, then updates motion paths to deliver the fastest and most precise way to execute on the given task. With just one more tap, the RMO starts the new job.

That’s all there is to it. No support tickets. No systems integrators required. Smart Setup handles everything automatically using built-in computer vision, running on Rapid’s edge-computing platform. 

Now here is the trick, every application and part is different, including different tolerances and requirements. There is no silver bullet technology that solves all the edge cases out there. That's why we qualify every deployment, to make sure the technology is a good fit for the application at hand, and there are no surprises or bitter taste after a robot installation. 

There’s been a lot of talk about pilot purgatory for AI projects. Companies get a solution running in a lab or a small pilot. But bringing it – at scale – into the real world can be a challenge. How do you overcome this?

Being realistic with the metrics you want to hit from day one, and defining a path to deployment with gateways along the way that unlock automatically the next stage. Wishful thinking is not a path. If something works in a lab, it does not guarantee at all that it will work in production.

Implementing AI in your production environment should not just be about getting the coolest new technologies and being on the latest hype. The use of AI needs to be firmly anchored in improving manufacturing outcomes.

Scaling is key, and should be considered when looking into AI applications in manufacturing. You don’t want your precious R&D team to work and solve a challenging problem, only to realize you cannot afford to maintain it in operation, or that the infrastructure would be too costly. And in that case, the technology is stuck again in a new purgatory. 

AI only for the big players? How do small and medium-size companies take advantage of these technologies? How do we democratizing the deployment of AI and smart automation? 

This is a "game of numbers". If you are an SME and have a couple of robots, you may not be able to get the full advantages of AI. You simply won't have enough data. If you have a fleet of hundreds of robots, then interesting trends arise. 

That’s why Robotics as a Service is such an interesting model for robotics. In those scenarios, the company providing the robot operator has the visibility of hundreds of systems, can use the data to improve the algorithms centrally, and every deployment benefits from the shared wisdom. 

The same applies to smart automation. The majority of SMEs out there don’t have any robots! Either they are too expensive or the risk is too high for them. Let me put an example of a success story: Westec Plastics corporation. They are one of the Bay Area’s best-known and most-respected manufacturing operations. As many other manufacturers, they are facing a labor shortage problem and conventional automation was too expensive and inflexible for their high-mix environment. Back in early 2020, Westec had a new contract to mold plastic PCR ‘chips’ for COVID-19 test kits. We (Rapid Robotics) worked with them and they “hired” a rapid machine operator (RMO) for the test-kit production line. In less than a week, the RMO was working on the line, neatly cutting the plastic gates between trays that came out of Westec’s four-cavity injection molding machine. Problem: Westec’s customer placed an order for 10 million pieces — but canceled partway through the job! With conventional robots, the cancellation would have been a costly fiasco for Westec’s fledgling automation program. Instead, the RMO was repurposed and carried on with different tasks, helping Westec human operators pad-print everything from medical devices to thermostat housings. In this task, a human operator might print one side of a part and stage it into a holding fixture. The robot would then take the part from the fixture,stamp it with the pad printer, pick it back up and load it into a drying tray. When the tray was full, the RMO would push it into a bin for bulk packaging. a human operator might print one side of a part and stage it into a holding fixture. It would then take the part from the fixture,stamp it with the pad printer, pick it back up and load it into a drying tray. When the tray was full, the RMO would push it into a bin for bulk packaging. 

We hear a lot about the semiconductor shortage, but that’s just the tip of the iceberg. Contract manufacturers can’t produce gaskets, vials, labels. We’ve seen cases where the inability to produce a single piece of U-shaped black plastic brought an entire auto line to a halt. SMEs are the backbone of this country and we need to empower them with automation to make sure they are successful. That requires different approaches to automation. Big system integration budgets and, difficult to find and costly to hire, AI engineers don’t fit the SME needs. They need the benefits of a fleet of hundreds of robots even if they only own a few of them. And that’s why companies like Rapid Robotics exist; to bring AI and automation to the 98% of manufacturing that currently cannot find labor and is skeptical of traditional approaches to robotics.

For SMEs reading this, if you are still manually performing machine-tending tasks, such as injection molding, pad printing, heat stamping, lathe tending, etc. it is time to give automation a chance! 

Juan is a robotics and automation enthusiast with the goal to scale and democratize the access to robotics technology in manufacturing and beyond. During his career, he has brought together the worlds of Industrial Automation, Robotics and AI. His work has been featured at the New York Times, MIT Tech Review, Wired, Forbes, TheRobotReport and other media outlets. He is the VP of Product at Rapid Robotics, where he is working towards making automation scalable, accessible and affordable. Prior to Rapid, Juan was the VP of Product at Ready Robotics, where he launched Forge/OS 5, an award-winning industrial platform for robots and automation. Before that, Juan was the Head of Advanced Manufacturing Automation for Siemens, where he led a top team of researchers and engineers in Berkeley, California, working elbow-to-elbow with renowned academic partners in the area of robotics and AI. Juan is a Technical Advisor for the Advanced Robotics in Manufacturing (ARM), member of A3’s AI Tech Strategy Board and Skydeck advisor. In 2019, he was awarded the MIT Tech Review Innovator under35 Europe in the Pioneer category. In 2020, he was awarded Siemens Inventor of the year and the prestigious Thomas Alva Edison Patent Award.