Brilliant Products

This Nimble-Fingered Robot Has a Grip that Rivals Yours

From arcade claw machines to prosthetic hands, robots aren't exactly known for having a delicate touch. That lack of dexterity has proven to be a real challenge in robotics, puzzling engineers and has derailing some important technological advancements. That's why DexNet 2.0 is so impressive: it's a robot that has super-precise grasping abilities that could revolutionize the future of manufacturing.

The robot’s deep-learning system sees a new object and figures out how to grasp it.

Get a Grip

The human hand naturally has two ways of grasping objects: power grip and precision grip. Power grip refers to the way we grasp and move heavier objects such as door handles and large bottles. Precision movement refers to how we grasp and direct small objects such as a pencil or sewing needle, a process largely controlled by the brain.

These natural functions provide the basis for how robots are designed when it comes to movement. While machines have been used to grip and move objects in industrial settings for more than 100 years, they had to be designed for very specialized tasks. Even then, their movements are rough compared to human workers—think of a candy factory where machines are used for certain tasks like cooking and shaping, but the delicate work of sorting and wrapping the finished product is done by people. In in the 1960s, the HandyMan and later the Stanford arm, with its controllable "two-finger" parallel gripper, set the precedent for gripper technology through the 1980s. Technologies such as suction grippers were developed during that time, as well. In the 1990s and 2000s, gripper technology became more advanced, being used for a wide range of functions from NASA space missions to instruments for performing complex medical operations. Still, until until DexNet 2.0 proved to have a 99 percent success rate in picking up and moving everyday objects, no robot had truly rivaled human function.

Grasping the Future

DexNet 2.0 combines traditional approaches to robotics with new technologies such as deep learning to power its dexterity. The engineering team behind the devolpment is made up of Berkeley professor Ken Goldberg, researcher Jeff Mahler, and the Laboratory for Automation Science and Engineering, a.k.a. AUTOLAB. They fed a neural network—connected to a 3D sensor and robotic arm—a database of millions of three-dimensional shapes, which it quickly learned. That way, when DexNet 2.0 sees an object, the neural network quickly identifies its shape and uses the information to select the correct grasp. DexNet 2.0's success has already blown away academics and industry professionals, and may soon lead to new commercial products and advancements in artificial intelligence.

Watch And Learn: Our Favorite Content About Robots

DexNet Has 99% Precision Grasping

Precision Robot Peels A Grape

Written by Jamie Ludwig June 29, 2017

Curiosity uses cookies to improve site performance, for analytics and for advertising. By continuing to use our site, you accept our use of cookies, our Privacy Policy and Terms of Use.