Artificial Intelligence makes sorting solutions smarter
Covariant co-founder, President and Chief Scientist, and Professor of Engineering and Computer Science at University of California, Berkeley
A robot is a good alternative to repetitive work. But what happens when you add artificial intelligence? You get a machine that learns by itself and can take over physically demanding work. That is why bpost pioneers with the technology of Covariant, Flemish top scientist Pieter Abbeel’s company.
Pieter Abbeel is a Professor of Electrical Engineering and Computer Science, Director of the Berkeley Robot Learning Lab, and Co-Director of the Berkeley AI Research (BAIR) Lab at the renowned University of California, Berkeley. He is a global authority on artificial intelligence (AI) and also worked as a research scientist and advisor at Elon Musk’s OpenAI lab. ” My three co-founders and I started Covariant in 2017 with the mission to bring AI to the real world. Two and a half years later, our company has grown to 40 employees and we’ve seen great progress as we’ve developed and deployed AI Robotics stations in the real world.
Abbeel, who lives just outside Silicon Valley, regularly meets with top business executives all over the world, giving them briefings on the latest research and the future of AI. During one of those working visits, he met the CEO of bpost, who introduced him to Marc Sorgeloos, Manager Automation & New Equipment at bpost. They teamed up to figure out the best opportunities at bpost for AI Robotics solutions.
“Our activities are vertically structured. We make parcels, collect them, sort them, take care of distribution and ultimately deliver them to the customer. And we do that in a mature market with a high standard of living. It is the ideal context to automate. With traditional sorting machines alone, we cannot make it, because we will meet our limits. That is why we work together with a specialist in AI-robotics technology in the logistics sector.”
“We’re building the Covariant Brain: universal AI that allows robots to see, reason and act on the world around them,” says Abbeel. “Our software powers all different kinds of robotic stations and can handle a wide range of tasks and items, including parcels and packages that come into bpost’s facility every day.”
bpost installed a Proof of Concept with a robotic arm in its sorting center in Antwerp. “We had identified four challenges,” says Marc Sorgeloos. “It started with removing parcels from our containers and placing them on the input side of the sorting machine, just like an employee would do it. In other words, not just grab it, but perform the right action depending on the parcel. After that, MARS – multipurpose AI robotics station – had to do the same with China rest mail, all kinds of small parcels that mainly come from China, are often wrapped in recycled plastic and on which all kinds of stickers are glued. “
“Powered by AI, robots can learn to see, pick and place any type of item, from parcels to packages to common household goods”
The next challenge for MARS? “Handling mail items, including magazines without wrappings. The robothad to recognize them so that he would not tear them to pieces. That, too, was a huge success, without changing in any way the setup. Covariant only had to teach the underlying technology about postal items.”
The final challenge consisted of common retail goods such as shirts wrapped in plastic polybags, pill bottles and boxes of bandaids. “Covariant’s technology worked well on every item, regardless of size or shape. The same technology was able to work on every type of item. We were so impressed that our subsidiary Active Ants placed an order almost immediately (see insert).”
Why were these such huge challenges? “Real world environments like bpost’s facility are challenging for robots because there’s an unlimited number of scenarios they encounter. Bpost receives millions of packages and parcels each day, and no two are exactly alike. It’s impossible to program a robot to handle millions of objects — that’s why you need AI. Powered by AI, a robot can figure out how to handle each unique item without being told what to do,” says Abbeel. “Through our work with bpost, we’ve been able to bring the most advanced AI research to these stations, together mastering the challenge of picking a wide variety of letters and parcels.”
Sorgeloos confirms: “Covariant supplies the most advanced technology in the world, with which we can relieve our employees of stressful and heavy work.” During the demos, employees and trade unions were also allowed to take a look. “We told them the whole story: with this technology in combination with an adapted robot you can start automating the heavy manual work and thus avoid back and shoulder problems. In the long term, you can compensate for the increasing shortage of labor in a rapidly growing parcel market. In addition, visual AI technology also comes into view. It checks and monitors the flow of parcels, so that it can prevent and correct any errors during the process, “says Sorgeloos. “The bottom line is that this technology allows you to continuously monitor and adjust complex machines and processes meticulously.”
“With this technology you can start automating the heavy manual work and thus avoid back and shoulder problems”
Abbeel is very enthusiastic about bpost’s cautious approach. “Compare it to self-driving cars. You do not immediately test it on the highway, but you move ahead step by step. “