February 12, 2013
Man And Machine: Human And Robots Team Up In Cross Training Technique
Lee Rannals for redOrbit.com - Your Universe Online
Scientists are investigating whether studies designed to help train people could also be applied to teams of both robots and humans, in a technique known as cross-training.MIT researchers will be presenting a paper at the International Conference on Human-Robot Interaction in Tokyo about the results of experiments they carried out with the cross-training technique.
Researchers have been undertaking various efforts to try and establish that robots and humans can operate safely side-by-side. However, more studies need to be performed in order to ensure robots are smart enough to work effectively with people.
Julie Shah, an assistant professor of aeronautics and astronautics at MIT and head of the Interactive Robotics Group in the Computer Science and Artificial Intelligence Laboratory, said there is a mismatch between the way robots are programmed to perform tasks and when they are programed to work in concert with people.
Past experiments involved a human trainer who would give a positive or negative response each time a robot performed a task. However, Shah and her colleagues´ new technique allows people to form a better idea of how their role might affect the robots.
MIT researchers designed a new algorithm to allow the devices to learn from their role-swapping experiences by modifying existing algorithms to allow the robots to take in information from positive and negative rewards, as well as data gained through demonstration.
By watching the humans switch roles to carry out their work, the robots were able to utilize these algorithms and learn how the humans wanted them to perform the same task.
The researchers set up the human-robot team to carry out a simulated task in a virtual environment, with half of the teams using the conventional interactive reward approach, and the other half using the cross-training technique. Once the teams completed the simulated tasks, they were asked to carry them out in the real world.
Cross-training teams saw a 71 percent increase when working at the same time after the training. Humans who spent time doing nothing, while waiting for the robot to complete a stage of the task, decreased by 41 percent.
The MIT researchers also found learning algorithms recorded a much lower level of uncertainty about what their human teammate was likely to do next. Additionally, human participants were more likely to say the robot had carried out the task according to their preferences than what those in the reward-only group reported.
“This is the first evidence that human-robot teamwork is improved when a human and robot train together by switching roles, in a manner similar to effective human team training practices,” said Stefanos Nikolaidis, a PhD student on the team.
According to Shah, the improvement in team performance could be due to the greater involvement of both parties in the cross-training processes.
“When the person trains the robot through reward it is one-way: The person says ℠good robot´ or the person says ℠bad robot,´ and it´s a very one-way passage of information,” Shah says. “But when you switch roles the person is better able to adapt to the robot´s capabilities and learn what it is likely to do, and so we think that it is adaptation on the person´s side that results in a better team performance.”