Humans and robots work better together if they can swap roles and learn from each other, a finding that may pave the way for the use of more robots on assembly lines, in space, on farms or in any kind of dangerous, dirty or boring operation.
“Cross-training,” or switching jobs, improves efficiency, as well as robots' confidence and the trust of human masters, researchers at the Massachusetts Institute of Technology found.
“We want robots to work alongside people,” said Julie Shah, assistant professor of aeronautics and astronautics at MIT and head of the Interactive Robotics Group in the Computer Science and Artificial Intelligence Laboratory.
“There’s a trend of removing the cages around robots and integrating them with manual work. That’s coming around with greater computation and sensing. We don’t want the robots to do the same task over and over again. They can’t if they are going to work in this dynamic environment. What we're interested is how we can make them smart enough to assist people.”
Shah and graduate student Stefanos Nikolaidis set up a simple experiment that involved 36 people coming into a lab and working with a robot arm to drill three screws into a board. The actual drilling was simulated. Just like an assembly line, they placed the screws, the board moved down the line and the robot pressed in the fasteners.
Shah said the volunteers all had different ways of working with the robot arm. “Some people didn’t like the robot; they were nervous,” she said. “They preferred to put in all three screws, stand back, and watch the robot drill. In other cases, people wanted to do it as fast as possible, so the robot drilled as fast as the screws were placed.”
By switching job roles, the robot was able to learn more quickly and the human began to trust the bot more. This human-robot cross-training worked better than giving positive or negative rewards to the robot.
“Cross training results in better teamwork,” Shah said.
The other big issue is forging trust between the two, according to Nikolaidis. This is important if the job involves tasks such as lifting heavy objects or performing dangerous work.
“If a person has learned to do something for 20 years, it’s hard to change and to place trust in a machine,” Nikolaidis said. “It’s one of the biggest barriers we are facing.”
The MIT researchers say for robots to become more like helpers who can take and perform commands than simple lifters or drillers, they will have to be able to learn the way humans do. And that means new kinds of expectations for both members of the human-robot team.
“From a teamwork perspective, the way we train robots today looks nothing like how medical or military teams train,” Shah said. “If we can make a robot train together with a person, the human-robot team will work better.”
A paper detailing their work will be presented at the International Conference on Human-Robot Interaction in Tokyo in March.