Inquiry about Proprioception and Action Space Dimensions in Bigym Environment

[xuai05]

I am a user of the Bigym environment, and I noticed that the proprioception dimension in the observation space is set to 58. According to the paper, this seems to be due to the use of a floating base. I have a few questions regarding this, specifically about how these remaining 58 dimensions map to the robot’s control space.

Specifically, I would like to understand:

How do these 58 dimensions correspond to the robot's control space? For instance, which dimensions correspond to the control of the left arm, and which to the right arm, etc.?
Are there any additional materials or documentation available that provide further clarification on this mapping?

Additionally, I observed that the action space consists of 15 dimensions. Could you please provide more details on how these action dimensions map to specific robot control commands? If there are any related documents or explanations available, I would greatly appreciate them.

Thank you for your time and assistance, and I look forward to your response.

[chernyadev]

Hi @xuai05,

We've added a document describing the current proprioception and action spaces. Hopefully, it helps:
https://github.com/chernyadev/bigym/wiki/Action-and-Observation-Spaces

Feel free to ask any clarification questions!