Skip to content

refactor: allow training on any supported data type#570

Open
aditya-neuraco wants to merge 1 commit into
mainfrom
fix/allow-training-on-individual-datatype
Open

refactor: allow training on any supported data type#570
aditya-neuraco wants to merge 1 commit into
mainfrom
fix/allow-training-on-individual-datatype

Conversation

@aditya-neuraco
Copy link
Copy Markdown
Contributor

Features

  • Allow training on any data types, removes image and state coupling.

Items

@aditya-neuraco aditya-neuraco self-assigned this Apr 23, 2026
@aditya-neuraco aditya-neuraco added the version:minor non-breaking feature updates, new functions or endpoints label Apr 23, 2026
@github-actions
Copy link
Copy Markdown
Contributor

Consider updating changelogs/pending-changelog.md with a summary of this change for the release notes. This is optional and non-blocking.

1 similar comment
@github-actions
Copy link
Copy Markdown
Contributor

Consider updating changelogs/pending-changelog.md with a summary of this change for the release notes. This is optional and non-blocking.

@aditya-neuraco aditya-neuraco force-pushed the fix/allow-training-on-individual-datatype branch from 668a6d6 to efe28e1 Compare April 28, 2026 12:59
@aditya-neuraco aditya-neuraco marked this pull request as ready for review April 28, 2026 13:22
Comment thread neuracore/ml/algorithms/cnnmlp/cnnmlp.py Outdated
)

all_proprio = torch.cat(proprio_list, dim=-1) # (B, actual_state_dim)
if self.proprio_normalizer is None:
Copy link
Copy Markdown
Contributor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

I find it confusing, in line 357 it can be none

Tuple of (images, masks) where images is a list of tensors
[B, C, H, W] per camera and masks is a list of [B] tensors.
"""
if DataType.RGB_IMAGES not in self.input_data_types:
Copy link
Copy Markdown
Contributor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

If rgb_images are not in the inputs, you can simulate a single all black image otherwise pi0 will raise an error

Tuple of (images, masks) where images is a list of tensors
[B, C, H, W] per camera and masks is a list of [B] tensors.
"""
if DataType.RGB_IMAGES not in self.input_data_types:
Copy link
Copy Markdown
Contributor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

same as pi0

@aditya-neuraco aditya-neuraco force-pushed the fix/allow-training-on-individual-datatype branch 3 times, most recently from f0b645c to ad865ad Compare May 8, 2026 16:49
@aditya-neuraco aditya-neuraco force-pushed the fix/allow-training-on-individual-datatype branch from ad865ad to 2807c3b Compare May 8, 2026 19:07
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

version:minor non-breaking feature updates, new functions or endpoints

Projects

None yet

Development

Successfully merging this pull request may close these issues.

2 participants