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

Func compute_observations bug #61

Open
Huisouan opened this issue Feb 25, 2025 · 4 comments
Open

Func compute_observations bug #61

Huisouan opened this issue Feb 25, 2025 · 4 comments

Comments

@Huisouan
Copy link

When I tried to use func compute_humanoid_observations.It seems that 'GenesisSimulator' object has no attribute 'key_body_ids'.Which should be attribute of env.

Image

@Huisouan
Copy link
Author

Plus,if bucket_width <= 1 there will be CUDA error

Image

@tesslerc
Copy link
Collaborator

Thanks for the feedback.
I see the first issue indeed occurs when the observations are not set to max_coords_obs. We haven't used that configuration in a while.
You can fix by changing to self.env.key_body_ids. I'd expect some mismatch in the expected observation size that would need fixing in the config.

Regarding the prioritization buckets, we have removed this logic in the latest v2.1 update.
While this provides some more granularity in prioritization, it is very tricky to tune and may cause unwanted behavior in some edge cases.
We now use the periodic evaluations to set the motion priorities. This should be more stable and prevent the crash you experienced.

@Huisouan
Copy link
Author

Thank you for your response. I noticed that when the configurations of obs and ref_obs differ, the history_buffer encounters a tensor shape mismatch issue during reset. Perhaps the history_buffer could be initialized differently based on whether a dataset is used, with a separate buffer specifically designed for datasets. I am really looking forward to the 2.1 version update. Could you provide any information on the update schedule?

@tesslerc
Copy link
Collaborator

v2.1 was released yesterday with the motion-prioritization fixes.

If I understand correctly, the issue you're encountering is during reset_hist_ref since this line is hard coded to use the max-observations. Correct?
Seems like build_self_obs_demo is also hard-coded to the same max-obs format.

The fix seems to be making it all configurable. Need to dig into the various dependencies to see which of these must have the same format and which can be independent.

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

2 participants