I am a beginner in RL. After reading your article, I have the following doubts.
1. The naive question.
- As there is relabeling operation in the framework, what about the difference of trained policies between the scenario that the number of objects is large and the scenario which holds a few objects?
2. The possible typos in the paper.
- Page 3: double "the". "In this stage, we aim to further improve the performance of our pushing and grasping policies in cluttered environments by alternately training to relieve the the problem of distribution mismatch"
- Page 5: double "the". "whether alternating-training can relieve the the problem of distribution mismatch."
- Page 6: misspelling of "condition". "Goal-condtioned VPG"
- What's more, some sentences may be more clear with some modification, such as "[4] and [5] design a hierarchical framework by evaluating whether is ready to grasp after current push step". → "the robot is ready to grasp after current push step" (It is just a personal suggestion.)
Hi, Ms Xu,
I am a beginner in RL. After reading your article, I have the following doubts.
1. The naive question.
- As there is relabeling operation in the framework, what about the difference of trained policies between the scenario that the number of objects is large and the scenario which holds a few objects?2. The possible typos in the paper.
- Page 3: double "the". "In this stage, we aim to further improve the performance of our pushing and grasping policies in cluttered environments by alternately training to relieve the the problem of distribution mismatch" - Page 5: double "the". "whether alternating-training can relieve the the problem of distribution mismatch." - Page 6: misspelling of "condition". "Goal-condtioned VPG" - What's more, some sentences may be more clear with some modification, such as "[4] and [5] design a hierarchical framework by evaluating whether is ready to grasp after current push step". → "the robot is ready to grasp after current push step" (It is just a personal suggestion.)