This Robo Hand Can Grab Things Just Like You. Elon Musk’s Open AI has built a robotic system that contrary to the traditional robotic systems can integrate into our world without breaking things when it picks up. Simulated robotic hand by researchers at Open AI learns to manipulate a block through trial and error and transfers that knowledge to a robotic hand in the real world means that it ends up “inventing” characteristic grasps that humans already commonly use to handle objects.
The researchers used reinforcement learning technique to teach the hand grasp things. The hand was doing random things and researchers started to give it a reward each time it did something to help it move toward the goal it actually wants to achieve.
Researchers also punished the robot when it did something dumb. Thus they made the robot learn not to do that sort of things. And with the passage of time and experiments with its abilities made it more and more versatile in grasping things like a block.
The system is workable only because of the many different worlds that have been built inside the digital world of this robot.and that is the reason it does not fail in simulated multiverse, because it can practice in lots of different realities slightly different from one another. If the robot trains in a single world, once it transfers to the real world random variables will confuse the hell out of it, but fortunately it is trained in a world with different simulation to avoid damage or devastation to the things.
The robot has five fingers and 24 degrees of freedom and is learning to use those fingers from scratch, through trial and error in simulation. The robotic hand goes about something called a finger pivot a bit differently. Humans typically pinch the block with the thumb and either the middle or ring finger, and they pivot the block with flicks of the index finger but the robot hand grip with the thumb and little finger instead.
Though this isn’t the first time researchers have trained a robot in simulation so a physical robot could adopt that knowledge, massive disconnect between simulation and the real world is the actual challenge.