No matter what you are developing a robot for, it needs to have the ability to walk on its own.
Now, the thing is, we have machines that can walk but there is still a lot of friction, as they cannot get up on their own, correct their postures, or find their way back to the track without human intervention.
Google's team tackled this problem with AI
In a bid to make robots navigate on their own, Google Robotics' team made tweaks to existing algorithms used for helping machines walk.
The work done by the company enabled a four-legged machine to teach itself how to walk forward/backward, and turn.
It performed all the actions without any external help, and that too in a matter of a few hours.
How the team did this?
To train the robot's algorithms, the team ditched virtual simulations and introduced the system in a real field.
The natural environment allowed the machine to quickly learn and adapt to uneven terrain, but it still needed human support.
So, the team restricted the territory of the robot and configured it to learn multiple moves at the same time.
This, combined with trial and error, resulted in perfect navigation
Ultimately, by using trial-and-error to navigate through different restricted environments, the robot learned the ability to walk seamlessly.
The team said it could walk on a variety of surfaces without requiring human assistance.
Plus, they believe these algorithms can be adapted to different robots, allowing engineers to make machines more capable in a range of fields.