Shape of obs_dict["pros_hand_qpos"] #255
-
When I use trained model to eval on the updated env, I got following error:
I notice that this might be caused by the shape of I guess the latter one is right, is it? Then how should I fix this error to make my trained agents compatible with the updated env? |
Beta Was this translation helpful? Give feedback.
Replies: 3 comments
-
Hi Xiaoben, Thank you for your question. We indeed changed the observation space and we understand the inconvenience that this may cause. This change is due to the fact that in the previous observation state, the object qpos is counted twice (one is object qpos, and one is accidentally attached at the end of MPL qpos). Hence, we would like to provide you with the script to convert your previous policy to be compatible with the new observation state. Are you using SB3? This script is used for sb3 conversion but can serve as a reference for all other RL implementations. This is what you need to do:
Please let us know if you encounter any issues along the way and we are happy to help. We used this method to convert our baseline, and the performance stays the same. Best, |
Beta Was this translation helpful? Give feedback.
-
An alternative solution that you can consider is simply applying a preprocessing wrapper to your policy, that repeats the missing observation, extending the vector to its original size. Namely you would need to append copy the first element of |
Beta Was this translation helpful? Give feedback.
-
Closing this for now. Please reach out to us if you have any further questions. |
Beta Was this translation helpful? Give feedback.
Hi Xiaoben,
Thank you for your question. We indeed changed the observation space and we understand the inconvenience that this may cause. This change is due to the fact that in the previous observation state, the object qpos is counted twice (one is object qpos, and one is accidentally attached at the end of MPL qpos).
Hence, we would like to provide you with the script to convert your previous policy to be compatible with the new observation state. Are you using SB3? This script is used for sb3 conversion but can serve as a reference for all other RL implementations. This is what you need to do:
edit_baseline
n…