In what could prove to be one of the most important moves ever made by humanity, scientists have taught our soon-to-be-robot-overlords not to stab us.
But rather than have it flailing around with knives and frozen baguettes they have used a clever for of algorithmic learning to teach it the best trajectory for scanning a potentially dangerous item in the presence of people.
Baxter, as the checkout assistant of the future is called, is a multi-jointed machine that can move more flexibly than a human.
By giving Baxter feedback about its chosen moves when scanning something it can learn how best to perform the task without spilling an innocent shopper's guts over the conveyer belt.
Ashutosh Saxena, assistant professor of computer science, said: "We give the robot a lot of flexibility in learning.
"We build on our previous work in teaching robots to plan their actions, then the user can give corrective feedback."
In tests with users who were not part of the research team, most users were able to train the robot successfully on a particular task with just five corrective feedbacks.
The robots also were able to generalize what they learned, adjusting when the object, the environment or both were changed.
The research was supported by the U.S. Army Research Office, a Microsoft Faculty Fellowship and the National Science Foundation.