rlmodule 0.1.7

Flexible reinforcement learning models instantiators library

rlmodule

Flexible reinforcement learning models instantiators library

Function approximators simple, but still strong. RNN - GRU - LSTM / SAC

Now it only supports skrl, but is intended to be library agnostic - in later expansion

try other algos shared separate model

How to run

Install rlmodule from local code

• Make sure you are in base rlmodule dict.

• Start virtual env.

python3 -m venv venv
source venv/bin/activate

• Install library from local code

pip install -e .

Note: sometimes installation may fail, if there is a run/ dir present, you may need to remove it (TODO: fix)

rm -rf runs

Deep policies: LayerNorm and Residual MLPs

rlmodule supports per-layer LayerNorm on MlpCfg and a SimBa-style residual MLP (ResidualMlpCfg). These work for both separated and shared architectures, with independent per-head granularity via the new OutputLayerCfg.head field. See docs/layer_norm_and_residual.md for motivation, references, and copy-pasteable cfg snippets.

Run chosen example

python3 rlmodule/skrl/torch/examples/gymnasium/pendulum_ppo_mlp_separate_model.py

Optional: observe run results in Tensorboard

tensorboard --logdir=runs/

Update new version to PIP

Change version name in pyproject.toml

pip install build twine

rm -rf runs
python -m build

twine upload dist/*