Advances in robotics have led to the creation of various devices that do a host of different tasks. However, too often these robots lack the means to do very specific tasks or understand what is expected of them.
Researchers at MIT’s Computer Science and Artificial Intelligence Laboratory (CSAIL) and Boston University have developed a feedback system that lets humans correct robot mistakes instantly using mind control.
The team used data from an electroencephalography (EEG) monitor that records brain activity allowing the system to detect if a person notices an error as a robot performs an object-sorting task. The machine-learning algorithms enable the system to classify brain waves in the space of 10 to 30 milliseconds.
The system could be used to help people who can’t communicate verbally via a series of discrete binary choices. It also brings the possibility of developing effective tools for brain-controlled prostheses.
“Imagine being able to instantaneously tell a robot to do a certain action without needing to type a command, push a button or even say a word,” says Daniela Rus, director of CSAIL. “A streamlined approach like that would improve our abilities to supervise factory robots, driverless cars and other technologies we haven’t even invented yet.”
Previously, EEG-controlled robots required training humans to “think” in a certain way that computers can recognize such as looking at one or two bright light displays, each corresponding to different tasks the robot could execute. The problem with this method is that it is taxing on a person’s thoughts, particularly for people who supervise tasks in navigation or construction that require intense concentration.
How It Works
The team instead focused on brain signals called “error-related potentials” (ErrPs), which are signals that are generated whenever our brains notice a mistake. When the robot decides on a choice it plans to make, the ErrPs determine if the human agrees with the decision.
“As you watch the robot, all you have to do is mentally agree or disagree with what it is doing,” Rus says. “You don’t have to train yourself to think in a certain way—the machine adapts to you, and not the other way around.”
Because ErrP signals are very faint, the system has to be fine-tuned enough to both classify the signal and incorporate it into the feedback loop for the human operator. The team also sought to detect “secondary errors” in the monitoring that account for when the system doesn’t notice the human’s original correction.
For example, if a robot is unsure about a decision, it can trigger a human response to get a more accurate answer. This improves accuracy and creates a continuous dialogue between the human and the robot.
The system is not yet able to recognize secondary errors in real time but when it can, researchers believe they can improve the accuracy to upwards of 90 percent. Future systems could extend to more complex multiple-choice tasks.