Implementations of Multi-Task and Meta-Learning baselines for the Metaworld benchmark
- Install uv
- Create a virtual environment for the project (with Python>=3.12)
uv venv .venv --python 3.12 - Activate the virtual environment
source .venv/bin/activate - Install the dependencies
uv pip install -e ".[cuda12]"
Important
Installing this package with no extras specified, i.e. uv pip install -e . will not work.
You need to be explicit about your choice of accelerator by specifying the correct extra when installing.
Note
The command in step 3. will install with NVIDIA GPU support. To use other accelerators, replace cuda12 with the appropriate accelerator name.
Valid options are:
cpu(No accelerator)tpu(GCP TPUs)cuda12(NVIDIA GPUs)metal(Apple Silicon)
For example, to install with TPU support, the proper commmand would be
uv pip install -e ".[tpu]"
Here is how you can navigate this repository:
examplescontains code for running baselines.metaworld_algorithms/rl/algorithmscontains the implementations of baseline algorithms (e.g. MTSAC, MTPPO, MAML, etc).metaworld_algorithms/nncontains the implementations of neural network architectures used in multi-task RL (e.g. Soft-Modules, PaCo, MOORE, etc).metaworld_algorithms/rl/networks.pycontains code that wraps these neural network building blocks into agent components (actor networks, critic networks, etc).metaworld_algorithms/rl/buffers.pycontains code for the buffers used.metaworld_algorithms/rl/algorithms/base.pycontains code for training loops (e.g. on-policy, off-policy, meta-rl).meatworld_algorithms/envsmetaworld.pycontains utilities for wrapping metaworld for use with these baselines.
