Google DeepMind unveiled robotic hands based on artificial intelligence

Engineers currently working on Google’s DeepMind project announced the development of artificial intelligence-based robotic hands.

One of them is called ALOHA Unleashed and the other is DemoStart, which is developed to enhance the capabilities of robotic hands that have multiple fingers, joints or sensors.

Most robotic hands that are built to pick up and move objects operate on their own without a second hand to assist them. In this new effort, the research team used artificial intelligence technology to teach a robot to use both hands to perform a difficult task, such as tying a shoelace. The result of this effort is called ALOHA Unleashed.

This robot is built on the new ALOHA 2 system and the ALOHA platform, which was developed at Stanford University in line with the development of remote applications. This new system improves dexterity and also allows two robot hands to be aware of each other as they work together on a common problem.

The robotic hands were trained through demonstrations to perform tasks such as hanging up shirts or making repairs. After that, the robotic hands achieved a degree of predictability to predict what their counterpart would do next.

According to this research team, achieving complex skills; Like using fingers, more joints and sensors in robotic hands require a degree of coordination between the two.

In fact, this coordination required the intervention of artificial intelligence in the learning process; With DemoStart, the team used reinforcement learning; This reinforcement learning involved giving simple tasks to the robotic hand and slowly increasing its difficulty. They found that they could teach a two-fingered robot with multiple joints and sensors to manipulate a Rubik’s cube, tighten a screw, and organize a workspace.