Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Contact forces are zero in Ant-v2 dataset #143

Open
PBarde opened this issue Feb 11, 2022 · 0 comments
Open

Contact forces are zero in Ant-v2 dataset #143

PBarde opened this issue Feb 11, 2022 · 0 comments

Comments

@PBarde
Copy link

PBarde commented Feb 11, 2022

I think this is related to this issue openai/gym#2593

I think that the Ant-v2 datasets where collected without contact forces. For example I found no observation with non-zero component between indexes 27 to 110 in ant-medium-v2. Is this intended or is it related to the issue linked above?

IQL agent stills learn reasonably well with this dataset (normalized return between 75 and 105) but is a bit unstable.

Is this expert score https://github.com/rail-berkeley/d4rl/blob/4aff6f8c46f62f9a57f79caa9287efefa45b6688/d4rl/gym_mujoco/__init__.py#L31 also derived without observing to the contact forces?

Thank you!

Config:
python 3.8.12
mujoco v2.1.0
gym==0.21.0
mujoco-py==2.1.2.14
dm-control @ git+https://github.com/deepmind/dm_control@4f1a9944bf74066b1ffe982632f20e6c687d45f1
d4rl @ git+ssh://[email protected]/rail-berkeley/d4rl.git@4aff6f8c46f62f9a57f79caa9287efefa45b6688#egg=d4rl

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Labels
None yet
Projects
None yet
Development

No branches or pull requests

1 participant