Contributing Guidelines
Contents
Contributing Guidelines#
Our team wholeheartedly welcomes the community to contribute to robomimic. Contributions from members of the community will help ensure the long-term success of this project. Before you plan to make contributions, here are important resources to get started with:
Read the robomimic documentation and paper
Check our latest status from existing issues, pull requests, and branches and avoid duplicate efforts
Join our ARISE Slack workspace for technical discussions. Please email us to be added to the workspace.
We encourage the community to make four major types of contributions:
Bug fixes: Address open issues and fix bugs presented in the
master
branch.Additional datasets and tasks: Make robomimic compatible with more kinds of datasets, simulators, and tasks.
New algorithms: Develop new algorithms or re-implement existing algorithms for learning from robot manipulation datasets.
Implement new functionalities: Implement new features, such as training with new kinds of observations (depth images, lidar, force/torque sensors).
Testing#
Before submitting your contributions, make sure that the changes do not break existing functionalities. We have a handful of tests for verifying the correctness of the code.
You can run all the tests with the following command in the tests folder of robomimic. Make sure that it does not throw any error before you proceed to the next step. Note that the tests can take a few minutes to run.
$ bash test.sh
Submission#
Please read the coding conventions below and make sure that your code is consistent with ours. When making a contribution, make a pull request to robomimic with an itemized list of what you have done. When you submit a pull request, it is immensely helpful to include example script(s) that showcase the proposed changes and highlight any new APIs. We always love to see more test coverage. When it is appropriate, add a new test to the tests folder for checking the correctness of your code.
Coding Conventions#
We value readability and adhere to the following coding conventions:
Indent using four spaces (soft tabs)
Always put spaces after list items and method parameters (e.g.,
[1, 2, 3]
rather than[1,2,3]
), and around operators and hash arrows (e.g.,x += 1
rather thanx+=1
)Use Google Python Style for the docstrings
For scripts such as in the
scripts
andexamples
folders, include a docstring at the top of the file that describes the high-level purpose of the script and/or instructions on how to use the scripts (if relevant).
Additional Coding Guidelines#
We also list additional suggested contributing guidelines that we adhered to during development.
When creating new networks (e.g. subclasses of
Module
inmodels/base_nets.py
), always put sub-modules into a property calledself.nets
, and if there is more than one sub-module, make it a module collection (such as atorch.nn.ModuleDict
). This is to ensure that the patternmodel.to(device)
works as expected with multiple levels of nested torch modules. As an example of nesting, see the_create_networks
function in theVAE
class (models/vae_nets.py
) and theMIMO_MLP
class (models/obs_nets.py
).Do not use default mutable arguments – they can lead to terrible bugs and unexpected behavior (see this link for more information). For this reason, in functions that expect optional dictionaries and lists (for example, the
core_kwargs
argument in theobs_encoder_factory
function, or thelayer_dims
argument in theMLP
class constructor), we use a default argument ofcore_kwargs=None
or an empty tuple (since tuples are immutable)layer_dims=()
.Prefer
torch.expand
overtorch.repeat
wherever possible, for memory efficiency. See this link for more details.
We look forward to your contributions. Thanks!
The robomimic core team