When it comes tasks like cooking, humans naturally know all the little steps that are required for doing something as simple as taking a cup out of the cabinet and filling it with milk. Robots on the other hand have to be taught all the little motions and steps like opening and closing cabinet doors, setting the cup down, opening the milk, pouring, and then closing the milk and putting it away.
Robots have to be taught each of those little steps; they can’t use intuition like people. Researchers at the University of Maryland Institute for Advanced Computer Studies are working on teaching robots to cook and they are using YouTube to do it. When watching these videos, the robots are able to learn grasping and manipulation motions needed for cooking by watching people on the videos.
The robots being taught with the videos use two recognition systems to process how the person in the video grabs something and uses it. The bots use artificial intelligence to process the information seen and language parsing to help understand spoken commands and translate those commands into action.
Together those processes help the robot to gather individual steps from various videos and assign them rules according to how the bot is programmed. The robot can then put the steps in the correct order. The process works, the team of researchers show an average precision of 77% and an average recall rate of 76% for the grasping recognition module. The object recognition module is more accurate with an average precision of 93% and an average recall of 93%.