For now, to teach a robot to perform a task requires either directly coding the robot or trial-and-error tests or even hand holding. But very soon, it’ll be possible to perform that task like you would any other day.
MIT scientists have developed a system, Planning with Uncertain Specifications (PUnS), that will help robots learn complicated tasks when they’d otherwise stumble, such as setting the dinner table.
It will no longer be the usual method where the robot receives rewards for performing the right task, PUnS has the bot hold “beliefs” over a variety of specifications and use a language (linear temporal logic) that lets it reason about what it has to do right now and in the future.
Analysts said, “To nudge the robot toward the right outcome, the team set criteria that helps the robot satisfy its overall beliefs. The criteria can satisfy the formulas with the highest probability, the greatest number of formulas or even those with the least chance of failure.
“A designer could optimize a robot for safety if it’s working with hazardous materials, or consistent quality if it’s a factory model.”
The new MIT scientists system is much more effective than the former approaches in early testing. A PUnS-based robot only made six mistakes in 20,000 attempts at setting the table.
Even when the developers threw in complicated task like hiding a fork, the robot continued and finished the rest of the tasks and came back to the fork when it popped up. In that way, it demonstrated a human-like ability to set a clear overall goal and improvise.
The MIT scientists are working to enable the robot to not only learn by watching, but react to feedback. So you could give it verbal corrections or a critique of its performance, for instance.
Although that will involve much more work, but it says at a future where your household robots could adapt to new duties by watching you set an example is coming.