Commercialized robots, whether they’re being used in Amazon shipping warehouses or Tesla giga-factories, still have trouble picking up irregularly shaped objects. Robotic arms are typically relegated to moving cargo of similar size and have yet to require the skills to pick up oddly-shaped objects like shoes, open boxes, furniture, or recreational equipment. Until now, that is.
DexNet: The Most Nimble Robot Yet
Introducing DexNet 2.0 – a state-of-the-art robot designed by researchers at UC Berkeley to reinvent automation. Equipped with machine learning technology and the finger dexterity to grasp real-world objects at a 99% success rate, DexNet is poised to disrupt the industrial manufacturing industry.
Described in the Berkeley News as “the most nimble-fingered robot yet,” the DexNet 2.0 attaches to a robotic arm and comes with a built-in neural network of 6.7 million data points that recognizes thousands of irregular shapes. By creating an intuitive robot that relies on machine learning to grasp irregular shapes, Roboticists at UC Berkeley engineered an intuitive robot that recognizes one shape from the other. By not only accessing an abundance of information stored in a cloud but learning from it with real-world interactions, the sky is the limit for Berkeley’s robot.
UC Berkeley’s Research
Stephanie Tellex, an assistant professor at Brown University who specializes in robot learning told MIT Technology Review, “It’s hard to collect large data sets of robotic data. This research is exciting because it shows that a simulated data set can be used to train a model for gasping. And this model translates to real success on a physical robot.”
With a team that includes Berkeley professor Ken Goldberg, postdoctoral researcher Jeff Mahler, and The Laboratory for Automation Science and Engineering (AUTOLAB), the DexNet 2.0 prototype is certainly in good hands. “We can generate sufficient training data for deep neural networks in a day or so instead of running months of physical trials on a real robot,” says Mahler.
Robotics & Deep Learning Technology
With that being said, there is plenty of research left unfinished and trials still in progress before moving DexNet from the development phase to fully commercialized industrial robot. Thankfully, the UC Berkeley team is collaborating with Juan Aparicio – a research group head at Siemens in the hopes of making the transition from prototype to working robot a seamless one.
One thing is for sure, the researchers at UC Berkeley have paved the way for new developments in robotics and artificial intelligence. The collaboration between the two fields has yielded a more capable dexterous robot that is both 3x’s faster than its predecessor and moves irregular objects 99% of the time – something that has never been done before. “We’re producing better results but without that kind of experimentation,” says professor Ken Goldberg in an interview with MIT. “We’re very excited about this.”
Whether it’s the DOBOT M1, the uArm Swift Pro, or the DexNet 2.0, gesture-based robotic arms with precise repeatability rely heavily on machine learning. From clinical work inside hospitals to selecting the right products inside Amazon shipping facilities, robotics is the future of automation and will have a major impact on our supply chain management processes.