Google's Robot Learns To Walk Themselves Without Human Interactions

A team of researchers from Google Robotics in collaboration with professionals at the Georgia Institute of Technology has developed an advanced algorithm to provide four-legged robots with the ability to walk without any controls from humans.

>>> This Soft Robot Grabs Objects Just Like A Real Octopus

By using a deep reinforcement learning framework, researchers are able to integrate a safety-constrained framework, an automatic reset controller, and a multi-tasking learning procedure altogether. Instead of waiting for assistance from human operators when facing failures, the system will make use of them for improvements in the next time thanks to self-learning capabilities.

Interactions with humans are something unnecessary as a robot can learn a number of different movements and directions at the same time and effectively gain experiences in a restricted training space. For example, the robot will learn how to go forward, backward or turn right and left on various surface conditions like flat ground, soft mattress, and more.

Clever algorithms appear to be the most impressive innovation of this research, which results in lesser time but more effectiveness. After 15 hours of self-learning, it’s not difficult for the four-legged robot to walk across several terrain types smoothly. Researchers can start controlling this walking robot once they connect it with a gamepad.

According to Google researchers, a load of new algorithms is also in development for other types of robots and learning environments, aiming to better the robotics applying in all fields.

>>> Researchers To Develop Synthetic Skin That Feels Pain