Skip to content

ACDSLab/ACDS-Dynamics-Model-Learning

Repository files navigation

ACDS-Dynamics-Model-Learning

ACDS Dynamics Model Learning Repository. Integrates with the dynamics models that are learned for use in MPPI-Generic but is more general than that. The idea is to have a general repository that can be publicly for Jason Gibson's research using PyTorch. Generally this should be used to extract compact datasets from ROS bags.

Training A Model

Testing A Model

Running Test Cases

Activate the virtual environment first, then invoke pytest as a module (so the project packages import correctly):

source ~/venv/model_learning/bin/activate
python -m pytest tests/

Test files mirror the source tree under tests/ and the tree is fully packaged, so test-local imports use absolute paths (from tests.<pkg>.<module> import ...). Pytest configuration lives in the [tool.pytest.ini_options] table of pyproject.toml.

Slow tests

Tests marked @pytest.mark.slow are skipped by default and only run when you pass --runslow:

python -m pytest tests/                      # fast suite (slow tests skipped)
python -m pytest tests/ --runslow            # run everything, including slow tests
python -m pytest tests/ -m slow --runslow    # run ONLY the slow tests
python -m pytest tests/ --durations=25       # report the 25 slowest tests

To mark a test as slow, decorate the function or class with @pytest.mark.slow, or set pytestmark = pytest.mark.slow at the top of a module to mark every test in that file.

Conventions

Systems

The system defines the

  1. inputs
  2. outputs
  3. state
  4. additional states
  5. integration scheme

Input

Values used to input into the dynamics that are ground truth, typically controls of sensor readings.

Output

The final outputs of the models, typically acceleration values.

State

The propagated state of the system.

Additional / Complete State

Additional values that we need from the dataset in order to properly remove "bad" data. A good example of this would be ignoring all data that occurs in reverse that we can check based off of a gear topic.

Models

The model defines the parametric or learned model used to predict derivatives

Model Output

Specific outputs from the internal model

Creating A Dataset

About

A repository for the dynamics model learning research done within the ACDS lab. Specifically the work done in https://arxiv.org/abs/2605.12628

Resources

License

Stars

Watchers

Forks

Releases

No releases published

Packages

 
 
 

Contributors