It has been known that it's easier to program an Artificial Intelligence (AI) to talk than to walk.
If you want to reset a smart speaker, you just need to unplug it. But for robots, you can't just unplug it after falling from a cliff for it to stand up again.
To address this problem, the Google Brain team created a "forward policy" and a "reset policy".
These two are meant to deal with algorithms, telling an AI when it's about to do something that it can't recover from, and stop it from doing so.
According to a white paper submitted by researchers at the Google Brain team, "by learning a value function for the reset policy, we can automatically determine when the forward policy is about to enter a non-reversible state, providing for uncertainty-aware safety aborts."
Since robots aren’t really ready to walk on the surface of the Earth without humans controlling their acts, they're more like toddlers in which they can go anywhere and kill themselves.
And this research by Google Brain team represents a significant upgrade in the field of experimental robotics.
What the team did, is introducing automation into automation. The algorithm tweak gives robots foresight into something, so they don't require humans to reset them during learning sessions.
A deep learning network typically gains expertise at a task after learning and failing many many times.
For example, a robotic arm in a factory, through repetition, can do its job very well. This is called reinforcement training, and it’s powered by machine learning algorithms.
With the tweaked algorithms, Google is trying to eliminate the needs for humans to "reset" the robots if the AI fails in an experiment.
In short, this should enable AI to teach robots about what to do and be caution before attempting something dangerous..
