Valerie Varnuska, a resident of Westbury, NY, has an interest in machinery and robotics. Valerie Varnuska enjoys seeing the unique ways robots can walk and the developments made in the movement of robotic hands.
In a recent blog post written by Google Brain Team’s Sergey Levine, Timothy Lillicrap, and Mrinal Kalakrishnan, a plan was revealed for speeding up the process of robotic learning. The plan relies on sharing information among robots, which should ease the burden of teaching individual robots the same set of tasks.
By using cloud robotics combined with deep neural networks, it is expected that robots will be able to learn for themselves instead of being programmed. This will hasten the learning process and make it more effective, especially when complex tasks are involved. (Hopefully, they will build in from inception, a secure way for humans to have the capability of overriding any robot “decisions.”)
To support the plan, a series of studies was done that looked at the effectiveness of neural networks and cloud learning in three different scenarios. The first scenario taught robots basic motor skills through trial-and-error.
The robots attempted to open a door multiple times and then sent data about their performances to a server. This data built a new neural network that was updated and sent back to the robots.
Meanwhile, the second scenario created internal models of objects and behaviors in addition to the trial-and-error method. The third scenario involved humans teaching skills to robots. In all three tests, when data was shared among the robots, all the robots’ learning processes were accelerated.