Robots juggling is nothing new. In the past, they used paddles, cups, and other end effectors to do it. This robot from RAI Institute doesn’t use task specific passive end effectors. It uses feedback from onboard vision sensors to attempt various juggling patterns. This robot learns directly on hardware in real-time. As explained on YouTube, the robot learned the cascade in under 10 minutes. Skills that have common elements can be reused.
We have already seen robots like this take on baseball and other sports. This one can transition between cascade, half-shower, tennis, shower, and box patterns seamlessly.
[HT]