Rapid, efficient...And ambidextrous. The robots of the future come from California

DexNet-4.0_Sito.png

As we all know, the e-commerce sector is a rising industry. It might now even be facing an epochal turning point: robots have become ambidextrous, a new development which is bound to revolutionise the organization of many retail companies.
 
The team led by Jeffrey Mahler of the University of California, Berkeley, has developed a system of algorithms, called Dex-Net 4.0, which enables robots to overcome those prior limits that prevented them from using both arms and grasping a diverse range of objects.
 
The robotic systems that are currently used across major e-commerce centres use grippers with suction cups, which are obviously unable to grasp a diverse range of objects. The approach developed by these Californian researchers implies two different types of grippers - one with a suction cup, the other with parallel jaws - and is based on a reward mechanism: the robot quantifies the probability of success for each type of gripper and then quickly chooses which one to use. The two robotic arms can thus grasp, as ambidextrous, any type of object.
 
After applying the new programme to the robot, the research team trained the machine to grasp novel objects. Relying on the new system, the robot used both its robotic arms to pick more than 300 objects in an hour, with a 95% reliability level.
 
The results of the work have been published in the magazine Science Robotics and might soon give an impetus to e-commerce, thanks to much more efficient machines that can perceive objects and the surrounding environment (almost) like human beings.