A question about vision robotic manipulation task

[ShiguangSun]

Hi, I have a question here. In the Appendix C, 'We consider that the goal is achieved if both the arm and the puck are within 5cm of their respective target positions.' Is this calculated through two images of state and goal, or directly by reading groundtruth data of puck and arm？

[elliotchanesane31]

Hi ! Indeed, in the Multiworld environments, the rewards are calculated directly by reading the ground truth data.