Baxter is but a child, with bright eyes and a subtle grin. It sits at a table and cautiously lifts a can of spray paint, then dangles it over a box marked “WIRE.” The error seems to smack Baxter across the face—its eyebrows furrow and blush appears on its cheeks. It swings its arm to an adjacent box marked “PAINT” and drops in the can with a clunk and that spray-paint rattle.
“Good,” says a voice off-screen, as Baxter’s face reverts to a grin.
Baxter is in fact a robot, and an industrial one at that, with hulking arms meant for lifting much larger things than cans and wire. Its face is not flesh, but a screen. And its decisions are not entirely its own, but those of a human sitting across the table—a woman with electrodes strapped to her head. The setup detects a particular signal in her brain’s electrical activity when she sees a mistake. In real time, the woman telepathically scolds Baxter for choosing the wrong box, and the robot corrects.
Researchers didn’t set out to embarrass an innocent machine, but to push further into the frontier of human-robot interaction, as they detail in a paper published online today. More and more, you’ll be interacting with machines: You’ll share hospital corridors with robots delivering food and medicine, and you could even fly a plane with your thoughts alone. For the time being, though, interacting with robots is crazy-awkward—they’re stilted and, well, robotic. The challenge now is socializing them.
Today, communicating with the machines is mostly about typing or vocalizing commands, which creates lag time. Letting Baxter read your mind takes milliseconds. “It’s a new way of controlling the robot that I actually like to think of as being natural, in the sense that we aim to have the robot adapt to what the human would like to do,” says MIT roboticist Daniela Rus, a co-author on the study. Namely, don’t put the paint in the wrong box, dummy.
The underlying technology is shiny and new and complex, but the idea is straightforward. When you notice a mistake, your brain emits a faint type of signal, known in neuroscience as an error-related potential. But that’s among all the other electrical chaos coursing through your brain that an EEG picks up, so machine learning algorithms sniff out the signal. When Baxter is about to make a mistake, the system translates the error-related potentials in the woman’s brain into code a robot understands.
Read more:
https://www.wired.com/2017/03/baxter-robot-fixes-mistakes-reading-mind/
