English | EN ROBOTS IN COMPLEX ENVIRONMENTS
June 9, 2015
A robot to help you tidying your home, or one that helps you nurse your fragile grandma. To many this may still seem science fiction but according to professor Pieter Jonker of Delft University these scenarios will be real within twenty years from now. Mr. Jonker is developing robots that are designed to learn and train themselves to accomplish tasks in complex environments.
Most of today’s robots work in the industry, e.g. in large production plants where huge machines firmly bolted to the ground inexhaustibly are welding or putting parts in place. Such environments are extremely well defined and organized: everything is always on the same spot and all actions follow a precise and invariable sequence.
Everyday environments however are much more complex: think of your home, a hospital or a nursing facility. There robots are needed that are able to anticipate and adapt to changing circumstances. Of course you could try to imagine and capture every conceivable situation in a software program but in the end that would hardly be a feasible endeavour given the endless variety of possibilities and interferences in normal life. That’s why Mr. Jonker and many other researchers today are working on robots, which are able to learn from their experiences.more–>
Walking for instance. A legged machine is used from which the software is removed, so it is forced to really start from scratch. The robot is rewarded when a correct pace is being made while it is programmed for reaching a highscore. This machine is repeatedly fed with an amount of energy, so that it can make its moves. In the beginning, there will be a lot of “silly walks,” but after a while via the reward system the robot will repeat the winning steps. At the same time the machine will try out new ways of walking to achieve a higher score. Much in the same way robots can learn how to collaborate, for example in a soccer team.
About the author