1. This paper presents a learning model based control strategy for the cup-and-ball game, where a Universal Robots UR5e manipulator arm learns to catch a ball in one of the cups on a Kendama.
2. The control problem is divided into two sub-tasks: swinging the ball up in a constrained motion and catching the free-falling ball.
3. A novel iterative framework is proposed, where data is used to learn the support of the camera noise distribution iteratively in order to update the control policy.
This article provides an interesting approach to playing cup-and-ball with noisy camera observations. The authors present a learning model based control strategy for the cup-and-ball game, which divides the problem into two sub-tasks and utilizes noisy position feedback from an Intel RealSense D435 depth camera. The proposed iterative framework uses data to learn the support of the camera noise distribution in order to update the control policy, and guarantees that probability of catch increases as learned support nears true support of camera noise distribution.
The article appears to be well researched and presented in an unbiased manner, providing sufficient evidence for its claims and exploring counterarguments when necessary. All potential risks are noted, and both sides are presented equally throughout. There does not appear to be any promotional content or partiality present in this article, nor any unsupported claims or missing points of consideration.