Scientists are building an AI-powered robot to take apart your old phone

In 2016, Apple announced that it had developed a recycling robot, called Liam, that could deconstruct an iPhone in 11 seconds. Six years and several machine generations later, Apple still won’t disclose how many iPhones its robots have recycled for parts.

But the potential impact of artificially intelligent robots on e-waste recycling more broadly might soon become clear, thanks to a new research project that seeks to develop AI-powered tools that allow a robotic recycler to harvest parts from many different models of phones. If such technology can be commercialized, researchers are hopeful it could vastly improve the recycling of smartphones and other small, portable electronics.

While today’s e-waste recyclers are mostly handling larger legacy devices like CRT TVs, a growing number of smaller electronics like smartphones and tablets have started to reach their facilities. This creates new challenges, as these devices are often difficult and time-consuming to take apart. Instead of salvaging potentially valuable components like the motherboard, recyclers typically remove the battery and shred the rest. Precious materials are lost in the process, and all of the energy that went into manufacturing components needs to be expended again to create new ones.

For several years, scientists have been exploring whether artificially intelligent robots could streamline the recycling process, making the recovery and reuse of parts from dead consumer electronics more economical. In December, the idea received a high-level boost when the US Department of Energy awarded a $445,000 grant to researchers from Idaho National Laboratory, the University of Buffalo, Iowa State University, and e-waste recycler Sunnking to develop software that allows robots to automatically identify different types of smartphones on a recycling line, remove the batteries, and harvest various high-value components. By the end of the two-year research project, the team hopes to field-test an early version of its technology at one of Sunnking’s facilities — after which it may pursue additional funding to commercialize robotic smartphone recyclers.

A JACK-OF-ALL-TRADES VERSION

Amanda LaGrange, CEO of the St. Paul-based e-waste recycler TechDump, says that the work these researchers are doing is critical for improving the sustainability of consumer electronics, which contain valuable metals and minerals that today’s crude recycling processes don’t recover. “Finding ways, like these scientists are with robots, of trying to reclaim rare earth metals is so important,” LaGrange tells The Verge. “Also, my jaded self is not convinced it can be done at scale at this point.”

Indeed, applying robotics and AI to e-waste recycling is a fairly new idea, and there aren’t a lot of practical examples of it working. The best-known example is Apple’s much-hyped line of recycling robots, but only a few versions of these robots are out in the wild, they only work on iPhones, and their impact on Apple’s overall e-waste remains murky at best. A jack-of-all-trades version that could be installed at an e-waste facility processing dozens of different models of smartphones has not been commercialized yet. The new research project aims to show that such a robot is, at least, possible to develop.

Various research teams will take the lead on different robotic recycling capabilities. Researchers at INL will focus on developing methods for removing batteries from smartphones using a robotic arm. In parallel, researchers at the University of Buffalo and Iowa State University will identify higher-value components, like circuit boards, cameras, and magnets, that can be removed from dead phones using the same robots and find or develop hardware to do the actual smartphone surgery.

The robots don’t just need good hardware, but software that allows them to quickly recognize different phone types and look up their internal anatomy. For this part of the project, Iowa State University researchers and Sunnking will be developing a database that includes 2D images and 3D scanning data on various makes and models of smartphones. Using a machine learning approach, that database will train the software guiding the robots to locate the phone’s battery and high-value components and extract them.

“We’re going to train that system to look at phones and say, ‘This is an iPhone, this is a Samsung model XYZ,’ then go to a database and say, ‘This is where we’re going to cut the battery out,’” says INL’s Neal Yancey, the principal investigator on the project.

Eventually, the researchers hope to have a smartphone-stripping robot that can be plugged into existing e-waste recycling operations. Sunnking, which will be providing 100 samples of five different phone models for the researchers to experiment with, will be the first to test that system out toward the end of the two-year project window.

At the same time, researchers at INL will analyze the economics of the entire robotic disassembly process to determine if it actually reduces recycling costs. The team’s goal is to improve materials recovery by at least 10 percent and recycling economics by at least 15 percent compared with standard recycling operations today.

Even those seemingly modest goals may be difficult to achieve. Adding specialized robotic arms to e-waste operations where phones are currently taken apart by hand will require a potentially sizable up-front investment. (The cost of robotic arms can vary widely, but the popular UR5 series sell for upwards of $35,000 apiece.) And with most of today’s robots designed for simple, repetitive tasks rather than the precision work of removing tiny phone parts, developing a robot that can measure up to its human counterparts in terms of disassembly speed and accuracy is no small feat, says Minghui Zheng, a roboticist at the University of Buffalo and co-principal investigator on the project.

“There are lots of limitations of robots,” Zheng says. Basic tasks, like using robotic grippers to pull out small components, could be “very challenging,” she says.

Developing AI-based software tools that can sift through the complex mixture of dead devices in an e-waste stream and accurately classify them could also prove challenging, although similar tools exist for sorting through solid wastes like plastic. Other groups are also attempting to develop AI-based e-waste sorting methods, including Carnegie Mellon University’s Biorobotics Lab, which recently worked with Apple on one such project.

Even if the initial research is promising, more work will be needed before AI-powered robots are a practical solution for handling the estimated 150,000 tons of portable consumer electronic waste Americans produce each year (a figure including not just smartphones but tablets and wearables like Apple Watch). With the initial project focused on just five of the hundreds of smartphones out there, the tech will need to be developed further to be practical for most recyclers. To process large volumes of smartphones in an industrial setting, the system will also need to be scaled up.

Product design changes could create another barrier to robotic recycling. As companies tweak their devices year after year, recycling robots will need to be kept up to date with hardware and software capable of handling the latest models. An e-waste recycler that’s considering investing in such technology might reasonably worry that in 10 years, new phone designs will have rendered the robots obsolete.

That’s why it’s so important that recyclability is baked into product design, says Sara Behdad, a sustainable electronics researcher at the University of Florida who’s not involved with the new research project. While Behdad says that greater use of robots could improve e-waste recycling “a lot,” she believes that many of the issues plaguing recyclers today, from glued-in batteries to proprietary screws, should be addressed through design for disassembly standards.

Such an approach would mean “less uncertainty” for recyclers in the future, Behdad says. And taking phones apart would be “much more within the capabilities of robots.”

Source: www.theverge.com