The researchers at OpenAI have developed a self teaching algorithm which is giving robot hand significantly new dexterity dubbed as Dactyl. It is another big step in the area of AI-Robotics integration.
This creation of the researchers is able to teach itself to manipulate a cube with uncanny skills by practice of years inside a computer simulation. Initially, the human element was eliminated from all aspects of learning and the comprehensive system required some another elements.
But all these shortcomings have been completed by developing a robotic hand that could execute more realistic object gripping. This algorithm uses a machine learning technique which is known as reinforcement technique. This robotic hand was given a task of manoeuvring a cube.
Then it was left to find out through trial and error that these movements would produce the targeted results or not. It is being shown by the videos that it is rotating the cube with efficient agility. It automatically found out various grips which humans commonly use.
But the researches are also showing that the robot was able in manipulating the cube sufficiently after some practices. There were multiple key advantages in this robotic hand but a key advantage in the Open AI research was transferring the robotic hand’s software learning to the real world.
In addition to this, it is also overcoming the reality gap by the efforts of the researchers of Open AI between the simulation and physical tasks. For this purpose, the researchers injected some noise into the software simulation by way of making the robotic hand’s virtual world efficient enough which it was not befuddled by the unexpected in the real world.