Using a modelless AI algorithm, they created a simulated anthropomorphic hand that can manipulate more than 2,000 objects.
Usually when the hand picks up an object or object, then we do not need to think about how the hand will adapt when holding the object or object in our hands. This is a naturally occurring process and occurs over a short period of time. This process is also what allows children to be more dexterous with their hands than robot the most advanced that exists today, for the business of holding objects.
However, this could change in the near future. A team of scientists from the Computer Science and Artificial Intelligence Laboratory (CSAIL) WITH has developed a system that makes robots have the same dexterity. Using a modelless AI algorithm, they created a simulated anthropomorphic hand that can manipulate more than 2,000 objects. Moreover, the system does not need to know what to take and how to hold it perfectly.
Reported from Gizmochina (8/11), for the time being the system is not ready to be used in the real world. To get started, the team needed to transfer them to real robots. This may not be as much of a barrier as we thought. Earlier this year, researchers from Zhejiang University and the University of Edinburgh succeeded in transferring the AI systems approach to a robotic dog named Jueying. The system allows the robot to learn how to walk and stand from falls.
As is known, the current system is not perfect. But the system is said to already be able to handle a lot of smaller objects, including things like tennis balls and apples with a nearly 100% success rate. But tools such as screwdrivers and scissors can not be done well. In the case of such an object, the success rate dropped to close to 30%. However, the potential of this system is enormous. It could someday make robots better at manipulating tools and make them much more efficient at tasks like packing boxes.
–
