Rainbow Dash can walk on many surfaces such as foam mattress or foot mat with many random bends.

The field of robotics has been increasingly advanced, including a robot that carries on Rainbow Dash with the ability to learn to walk on its own. This four-legged robot only needs a few hours to learn to walk backward and forward, turn right and left.

Researchers from Google, UC Berkeley and the Georgia Institute of Technology have published a paper on the site  ArXiv describing a statistical AI technique called deep reinforcement learning that they have used to generate intelligent robotics achievement. new generation smart.

Most of the previous intelligent self-learning techniques took place in a computer simulation environment. However, Rainbow Dash used this technology to learn how to walk in a realistic physical environment. Furthermore, it can do so without the need for a dedicated teaching mechanism, such as an instructor or pre-programmed data. Rainbow Dash successfully walked on a variety of surfaces, including soft foam mattresses and foot mats with lots of random kinks.

The deep learning techniques that robots use include a type of continuous right and wrong machine learning by repeatedly interacting with the environment. This is similar to computer games that digitally learn how to play to win. This form of machine learning is markedly different from traditional supervised or unsupervised learning, where machine learning models require training data to be clearly delineated. Deep reinforcement learning combines reinforcement learning methods with deep learning, where the scale of traditional machine learning is greatly expanded by the power of huge computations.

Robots can self-learn and operate independently without needing an instructor or pre-programmed data. Photo: Techxplore

Although the team thinks that Rainbow Dash has learned to walk on its own, human intervention still plays an important role in achieving that goal. The researchers had to create boundaries, the robot had to learn to walk to keep it from leaving the area. They also had to devise specific algorithms to prevent the robot from falling, one of which focused on constraining the robot’s movement. To prevent accidents and damage from falling, robot reinforcement learning often takes place in a digital environment before algorithms are converted to physical form to protect the safety of the robot.

Rainbow Dash’s success came about a year after researchers found a way for the robot to learn the actual physical environment instead of the virtual one as before. “It’s really hard to remove humans from the robot’s learning process,” said Chelsea Finn, a Stanford assistant professor affiliated with Google. By allowing the robot to learn autonomously, it can operate closer to real-world augmented deep learning.”