A new robot developed by researchers at Cornell University isn’t only capable of assisting you with tasks, it can accurately predict when you might need a hand. Armed with a Kinect sensor, the robot — developed by Ashutosh Saxena and his team of computer scientists — utilizes a dataset of 120 videos to analyze and understand your movements. It can then help you perform certain tasks including making a meal, stacking or arranging objects, or taking your medicine.
To detect your actions, researchers use complex algorithms to detect skeletal movements that are assigned to a sub-activity, which include reaching, moving, pouring, eating, and drinking. Activities are associated with the objects around you, allowing the robot to assess future actions and possible destinations. It can also learn from mistakes, improving its precision in the process. The robot has already achieved accuracy of 83.1 percent for detection of high-level activities, outperforming the algorithms it was tested against.
via The Verge
