simpler-env
Coordinate Representations: SimplerEnv vs. Libero
If I use the postprocessing from this repo to run the Libero benchmark, the performance is much worse. I get that visuals are much more realistic here compared to Libero and that there is a difference between the low-level controllers.
However, I am still confused why all publications report successes around 80% and I barely get any observation right.
My Question: Does Simpler use the same coordinate representations as LIBERO?
- Angles are initially returned in Euler angles and then postprocessed to
axangle. Is this the same convention that Libero uses? - Do they use the same reference frames (i.e. the robot base frame)?
- Do both frameworks use a delta for xyz/angles or does libero use some absolute positioning?
I appreciate the work you put into this project!
This repo does not use any part from, nor does it have any relationship with, LIBERO. This repo uses maniskill envs and controller name for each robot is at https://github.com/simpler-env/SimplerEnv/blob/main/simpler_env/utils/env/env_builder.py#L20 . Controller for each robot can be found in https://github.com/simpler-env/ManiSkill2_real2sim/tree/ef7a4d4fdf4b69f2c2154db5b15b9ac8dfe10682/mani_skill2_real2sim/agents/configs .
