Contact Us

TEL: +86-755-27150460
Address: 2nd & 3rd Floor, Building 4, 4th Zone, Dejin Industiral Park, Fuyuan 1st Road, Heping Community, Fuhai Subdistrict, Baoan District, Shenzhen City, Guangdong Province, China.


Home > News > Content
When Can A Robotic Arm Be As Flexible As A Human Hand?
Aug 03, 2018

When can a robotic arm be as flexible as a human hand? “Previously, robots could only execute programs written by a team of engineers, but now they can learn more complex tasks themselves.” Although a robotic arm and five mechanical fingers are not as flexible as humans, in the world's top artificial intelligence laboratories, researchers are getting closer and closer to creating robotic arms that mimic real human hands.



In the OpenAI lab, created by Elon Musk and several other Silicon Valley celebrities, the researchers created a robotic arm called Dactyl. It looks a lot like the mechanical prosthesis of Luke Skywalker in the latest Star Wars movie: its mechanical fingers can bend or straighten like a human finger.

You can let Dactyl show you one side of the alphabet – like the red O, the orange P or the blue I – which will show you and then rotate, twist and flip the blocks in a flexible way.

 1 副.jpg

This is very simple for humans, but for a machine, this is a remarkable achievement: the robotic arm Dactyl is largely on its own to learn how to accomplish this task. Researchers use mathematics to let Dactyl learn, and they believe that robots and other machines can be trained to perform more complex tasks.



The system was created by researchers at Autolab, a robotic laboratory at the University of California at Berkeley, which represented the limits of technology a few years ago. The machine is equipped with a two-prong "plier" that picks up objects like screwdrivers or pliers and sorts them into different containers.

 2 副.jpg

Pliers are easier to control than five fingers, and the software needed to make a grip is not that difficult. It can handle some objects that are not familiar. For example, it may not know what a restaurant-style ketchup bottle, but it knows that the shape of the bottle is similar to a screwdriver. However, if the machine encounters something different than it had previously encountered - such as a plastic bracelet - it may not be handled very well.


Pick up

Everyone wants a robot that picks up anything, including something it has never seen before. This is the robot that other Autolab researchers have built over the past few years.

This system still uses simple hardware: a clip and a suction cup. But it can pick up a variety of random items - from scissors to plastic toy dinosaurs.

 3 副.jpg

The system benefits from the tremendous advances in machine learning. Berkeley researchers modeled the physical models of more than 10,000 objects and determined the best choice for each object. Then, the system analyzes all data using neural network algorithms and learns the best way to identify each item. In the past, researchers had to program robots to complete each task. But now, it can learn these tasks on its own.