A science teacher leads projects to achieve autonomy in their teams. Is it possible to get a robot change in behavior after a reprimand?
In principle, the scene may seem strange to imagine, but science is working hard to make, with shades, a reality. Professor of the Department of Electronics and Computer Science faculty Roberto Iglesias is one of those scientists who are devoting much of his existence to create a robot “smart” and know what to do when there is an unforeseen event. First of all, Iglesias explained that “although many people talk about it imagines something like the Star Wars, the robot is quite utopian ideal. This means that the fruits of exact science also make mistakes.
“All the robots, at a given time, they will make mistakes. But this is not the problem. The important thing is to be able to learn from them and recover without the need for a service technician to review, analyze and schedule for this, as well as impractical, would be expensive, “he said citing the example of a domestic robot collision with the wall and leave a mark or a line.
Negative reinforcement:
Removed the possibility of technical, Iglesias said that you can only get a change of behavior by a “negative reinforcement” or, in other words, with a punishment. “But of course, who punishes him? Y,. And above all, how do you do? “He quipped. This professor of the Faculty of Science has spent several years working on a system that allows this simple explanation in words materialize into something and, more importantly, something easy to handle. To that end, throughout 2009 he led a project of experimentation with low-cost robots in domestic environments, and now is leading another research on intelligent control in unknown environments, a project completed by the end of 2012 and that the Government has funded more than 160 446 more a Research Staff Training grant.
Trying to get his explanation was as informative as possible, Iglesias said the robots working in the USC move according to what they see. “Then we try to learn to see and do at once,” he said.
The robots have built-in laser scanners, sensors and computers, among many other pieces. To get them to change their behavior in terms of motion, for example, Iglesias explained that using a camera system that sends images to the robot, but only marks the paths you follow the people in that space, so that the appliance imitate and no crash.
“With this system, it takes seven to eight minutes to learn to go along without hitting the wall or away too much,” said researcher noting that, at USC, the research in robotics focuses on the ‘learning’ to achieving “autonomy” as similar to that of humans, which allows decisions to unexpected situations and in different environments, both known and unknown. The Star Wars is still fiction, but it requires less to make it happen.
