Babies are some of the fastest learners on Earth. This is in large part due to the low development of their prefrontal cortexes. Unlike adults with fully-developed prefrontal cortexes, children are able to perceive things through sound, smell, touch, taste, and sight that are not hindered by preconceived notions. For example: if an adult were given a stick, they would see a stick because that is what prior knowledge and experience tells them. A child could see a sword, a javelin, or flag pole to mark their new fort. What if we could teach robots to see and think as children do, learning new skills and having abstract ideas just as easily as a human baby?
There is a problem within robot and artificial intelligence programs that seek to teach robots what to do rather than tell robots what to do. Currently, programmers have to specifically code sequences that robots can use to work through problems and perform tasks. By creating a robot that can learn like a human does, the robot is no longer inhibited by its programming and can problem solve on its own. Computer scientists at the University of Washington saw an opportunity here and partnered with university developmental psychologists to help teach baby robots just like we teach baby humans.
The team of scientists and researchers published their theory in PLOS One. The overarching goal of the project is to have robots learn through their own experiences. By using infant research studies, they were able to build algorithms that mimic learning models. In the first experiment, they used gaze-based simulations to have a robot learn from a human purely based on sight. First, the baby bot watched a human move his head back and forth until the baby bot was imitating the human and looking at the same object as the human. The second part of the experiment involved the baby bot learning about blindfolds and that they stop humans from seeing. The robot learned not to look at where the blindfolded human was looking because it understood that the researcher could not see.
A mission of the project is for baby bots to learn more complicated functions or ideas from humans through experience. There will be other experiments based on other senses, such as touch and hearing that may be able to teach these robots faster and more effectively than traditional programming. Who knows, this could lead to the friendly and capable robots like R2D2, C-3PO, or the new BB-8.