In a blog post which has been released this week on OpenAI unveiled that the latest version of their OpenAI five learning algorithms that they have now wrapped into a system which is called as “Dactyl” which will let them train robots in different set of ways and also without having the physical based modelling. The result which can come in the combination with the Shadow Dexterous Hand, is a robot hand that can manipulate objects with near human like dexterity.
OpenAI has explained clearly that the Dactyl is trained in an environment within a virtual simulated and successfully adapts that acquired knowledge with respect to the real world.
At the time of doing different experiments, the researchers place a block in the hand of the robot and order dactyl to reposition it according to their choice. Dactyl then processes the hand fingertip coordinates and images from some of the three cameras which is being positioned all over around for examining it. The result that comes is just like the same as that of the eerily human like.
Robotics humanoid hands were first which were unveiled few years back but traditional robotics techniques have so far proved ineffective in coming up with the solutions to help them in manupilating objects efficiently so the Artificial Intelligence company, OpenAI has decided to use the domain randomization, which is a process created by the lab to solve the difficulties which is related to transferring simulated experiences into the real world experiences to as the reality gap to come up with a new set of approach.
Apart from just training a robot on a simulated environment, domain randomization exposes it to a several range of surroundings in a simulation which is designed with many set of experienced, and bingo that they have a winner.