Thinking robots learn about their world
While the more fanciful visions of 1950s sci-fi may not have been realized, robots have managed to establish themselves in controlled environments such as factories, or with controlled tasks like vacuuming. However, researchers from MIT in the U.S. say that robots can take the next step of carrying out complex tasks in previously-unknown locations by understanding their environments more accurately.
The key to achieving this is making robots more aware of their own limitations, say the researchers at MIT’s Computer Science and Artificial Intelligence Laboratory (CSAIL). This means the robots need to work out what they don’t know and take steps to find out more through further examination or asking questions.
The researchers have developed a system that uses something called a “state estimation component.” It calculates the probability of an object being what the robot thinks it is. If the robot is unsure what the object is, the robot will gather more information before taking any action.
“[The robot is] always thinking about its own belief about the world, and how to change its belief, by taking actions that will either gather more information or change the state of the world,” says Leslie Pack Kaelbling, the Panasonic Professor of Computer Science and Engineering at MIT. This means that robots don’t need to have a fixed plan about what they need to do for every situation. They can learn as they go along and make decisions on the fly, based on what they find out.