Robots can quickly and easily learn new things through crowdsourcing, says a new study done by a University of Washington research team. In other words, robots that use the internet to look up information are smarter.
“We’re trying to create a method for a robot to seek help from the whole world when it’s puzzled by something,” said Rajesh Rao, an associate professor of computer science and engineering and director of the Center for Sensorimotor Neural Engineering at the UW. “This is a way to go beyond just one-on-one interaction between a human and a robot by also learning from other humans around the world.”
In the past, robots have been taught new things by imitating humans. That is a slow process, though, as it takes many repetitive lessons to teach a robot anything. In a way, it is similar to teaching a dog a new trick.
Unlike a dog, however, robots can be programmed to use the internet to look up new information. If a robot learns the gist of a new skill, it can look up the more advanced parts of the skill online.
“Because our robots use machine-learning techniques, they require a lot of data to build accurate models of the task. The more data they have, the better model they can build. Our solution is to get that data from crowdsourcing,” said Maya Cakmak, a UW assistant professor of computer science and engineering.
The UW team devised a study that was able to show how robots could learn better with crowdsourcing.
They made a series of models out of Legos, including a simple model of a car, tree, snake, and turtle. They then asked their robot to build a similar models. The robot tried, but was ultimately unsuccessful in its endeavor to build models based on a few human-made ones.
The researchers then had the robot turn to the internet for help. They hired people on Amazon Mechanical Turk, a crowdsourcing site, to build over 100 models of each of the previously mentioned themes. The robot was then able to pick the best models based on difficulty, community rating, and more.
After that, the robot built numerous models for each of the themes it was given. Here are the models that the robot built for the “turtle” theme:
The robot ultimately came up with the best possible ways of achieving its goal – to build a turtle. This means that the models were generally less complex than the human ones, making them easier for the robot to build.
“The end result is still a turtle, but it’s something that is manageable for the robot and similar enough to the original model, so it achieves the same goal,” Cakmak explained.
The University of Washington team hopes to use crowdsourcing to teach the robot more new things, such as being able to fetch items from a multi-floored building. They hope to one day see robots using the internet more and more to learn new things and be able to assist humans better.