The baseball batting robots are two robot arms, where one is the pitcher and the other is the batter. The pitcher can target the strike zone at 40kph (25mph) 90% of the time, while the batter can hit the ball 90% of the time only 3.5 meters (11ft) away. In order to do this, the robots require cameras that capture the position of the ball at one thousand frames per second. What appears very fast to human eyes can be tracked slowly by the camera, which then informs the robot arms’ control software exactly where the ball is. In the future, the researchers expect the pitcher to throw the ball at 150kph (90mph), while the batter should be able to hit the ball in specific directions.
The baseball batting robot has been making the rounds through various news outlets since last week, and while the demonstration was cool, the object tracking computer vision software that allows the robot arms to perform the feat remains an unsung hero. Doing a little digging into the Ishikawa Komuro Laboratory (part of Tokyo University) website unearthed some cool videos that show off just how accurate their computer vision software really is.
For example the robot arms can not only pitch and bat balls, but can catch them at high speeds, dribble them so fast you can barely keep an eye on the ball, position themselves properly to shake a roving human hand, and tie knots with one hand. And while they haven’t perfected directional hitting, the robot can bat the ball into high or low nets repeatedly, even when the ball is thrown by a human pitcher. A different type of computer vision software tracks an object’s position in 3d space (videos after the break).
[source: Ishikawa Komuro Lab]