Artificial intelligence (AI) advancements are increasingly making headlines with the technology constantly evolving. However, the humanoid robots we have seen in many science fiction films are still far away from becoming reality.

Experts speaking at this year's global AI conference revealed this is because the industry is still struggling with certain key issues in closing the gap between humans and robots. One of these is the low efficiency in learning algorithms.It seems Elon Musk-founded AI research lab OpenAI may have just made a giant step in that direction. In a blog released on the organization's site this week, OpenAI revealed the evolution of their Open AI five learning algorithm into a system called Dactyl that can train robots without the input of physical-based modeling.

A robot with unprecedented dexterity :

Dactyl uses a Shadow dexterous hand to handle objects in a process that results in movements comparable to human mobility. "We’ve trained a human-like robot hand to manipulate physical objects with unprecedented dexterity," reads the non-profit's blog. 

OpenAI explained that Dactyl is trained entirely in simulation and successfully adapts that acquired knowledge into reality. The exercises see the researchers place a block in the palm of the robotic hand and order Dactyl to re-position it. Dactyl processes the hand's fingertip coordinates and images from three RGB cameras and almost elegantly twirls the block around. The resulting motion is eerily human-like.

Robotic humanoid hands were first introduced decades ago, however, traditional robotics proved ineffective in providing solutions for training the tools to manipulate objects efficiently. So OpenAI decided to use domain randomization, a process created by the lab to solve the difficulties related to transferring simulated experiences into the real world referred to as the reality gap.

Behaviors discovered autonomously :

Instead of training a model on a single simulated environment, domain randomization exposes it to a wide range of environments in a simulation designed with many experiences. The solution proved successful.