00:00You've played Jenga. Easy for you, but for a robot, it's a nightmare.
00:04Jenga is a very difficult problem for a robot because you have to rely on multiple different types of sensor
00:09input
00:10that comes from how hard you press the Jenga, how hard you squeeze it, where you should put it in
00:15the stack
00:16to avoid tipping the whole Jenga game over, which right now the robots are not really trained to do.
00:21That's why this San Francisco company is trying to teach them. Here's how.
00:25A worker plays Jenga wearing a camera on his head and sensor bands on his wrists.
00:30The camera sees what he sees. The wristbands pick up tiny electrical signals from his muscles.
00:35Put both together and you get a record of what the hand is doing even when the camera can't see
00:40it.
00:40The robot learns by imitating.
00:42You have the signals of both the signals that are coming through the muscles with how you are touching
00:47and moving the different bricks around, but also see the bricks in front of you
00:52so that you can coordinate where you should put it in to avoid tipping over the Jenga tower.
00:58And Jenga is just the test case. The company hopes this kind of data will help train humanoid robots
01:04and even self-driving cars and drones to act with a little more of the human touch.
01:09You have him?
01:10ända
